Haris Rauf, like a ham actor nicking the Oscar winner's scene, finishes it with a four off the first ball of the 48th over. The last Pakistan wicket has put together an unbeaten 89 to win the match for them. Rauf, the number 11, ends with 7 not out from 19 balls with one four, while Shadab Khan, soon to be named Player of the Match, has 113 not out from 96 balls with 13 fours and five sixes. Whatever mathematical possibilities might theoretically remain in practice England's tournament is over, since losing this match from the position they were in is the sort of body blow that no side can bounce straight back from.
#cricket #ODI #ODITriSeries #EnglandvPakistan #ENGvPAK
Search
298 results for “Five9”
-
-
FBI, the Secret Service, and German state investigators Landeskriminalamt Rheinland-Pfalz have been tracking these clowns since November 2024. The IOCs were RIGHT THERE. The TTPs were RIGHT THERE.
Pull the CISA advisory NOW, load those SafePay indicators of compromise into your monitoring stack, and check activity through August 2026 before someone calls a full wipe.
Reward: You've unlocked the Dunce Helm of the Unpatched — minus five to threat detection, minus ten to reading advisories. (2/3)
-
New Video of Big Storm Shadow Strike on Proton Plant in Oryol
by #Suchomimus
03:18https://www.youtube.com/watch?v=kwjEptFMNEM
In the Ukraine War, yesterday multiple Storm Shadow missiles hit the Proton Plant in Oryol, with at least four possibly five missiles hitting the facility.
🇬🇧 🇺🇦
Suchomimus gives regular updates on the Ukraine War, and the occasional dinosaur fossil! -
@lproven coattails? Exploitation of coolness?
Extremely unlikely, IMHO.
https://www.facesofopensource.com/jordan-hubbard-2/
https://www.linkedin.com/posts/johubbard_daemonarchy-111-is-released-httpslnkdin-share-7517486335977504768-hM2g/ five days ago (his response to a comment):
❮There's some context here, Scott Kamp - Omarchy is like any other desktop in a sense - you either like it or don't like it, and it's not like we didn't already have desktops going all the way back to Xcfe, KDE, Gnome, etc. The point wasn't to create "another streamlined desktop", the point was to show that the latest Buzzy Thing could be ported and wrapped up in a bow, using AI, in days instead of months. FOMO is history.❯
-
At one point I had four different pitch decks for the same piece of work. Same work, same cost, same engineers, four different stories, and not one of them led with the engineering argument, because that was the one that never worked.
Everyone I worked with agreed the tech debt was there. Very few of them wanted to do anything about it.
Twice I was brought in specifically to pay it down. One was a startup with a few years of debt and one explicit job, and deployments went from about eight person-hours to about twenty-five minutes. the other a retailer with decades piled into a monolith and a brief of about twelve words.
On the retailer, months went on paying interest before a penny came off the principal. From the outside that looks like nothing happening, and the name on the promise was mine rather than the company's.
You end up guarantor on a loan you never took out, for a lender who'll only lend if you over-promise.
#TechDebt #SoftwareEngineering #EngineeringLeadership #DevOps
-
Alpha School, the chain of AI-focused private schools promising pupils learn twice as much in two hours of classes, is missing its own targets. Internal documents seen by the New York Times show fewer than half of its pupils hit the 2x learning mark last year, one in five missed the reading goal and half did not learn maths twice as fast. https://gizmodo.com/the-ai-centric-private-school-chain-reportedly-isnt-meeting-its-own-academic-goals-2000830291 #AI #GenAI #AIEducation
-
A keyword sheet tells you what people type. It does not tell you how many pages you need, what goes on each one, or which ones you should not build at all. That gap is where most B2B sites lose time: 500 rows become 500 thin pages that all say the same thing, or 500 rows become one page that answers none of them.
This is the process for turning the sheet into a small number of pages that can actually ship — model pages (one part number, one page) and application pages (one use case, one page).
Step 1: Give the sheet the columns it actually needs
ColumnWhat goes in itkeywordExactly as typed, including the model number in the buyer's formatmarket / languageWhich language version this query belongs tointentOne of the five types in step 2page typeModel page, application page, category, technical notetarget pageExisting URL, or "to build"statusExisting / planned / merged into XnotesThe question behind the query, in one sentenceownerWho writes it
Most sheets have one column: keyword, maybe volume. Add these before anything else.
The two columns that do the most work are intent and notes. Notes is where you write the question in plain words: "replaces XYZ-120" actually means can I substitute this part without redesigning the board. That sentence becomes the first paragraph of the page. If you can't write it, you don't yet know what the page is for.Step 2: Sort every row into one of five intents
IntentQuery looks likePage typeIs it a landing page?Model lookup"XYZ-120 datasheet", "XYZ-120 price"Model pageYesCross-reference / replacement"replacement for ABC-45", "ABC-45 equivalent"Cross-reference page, or a block on the model pageYes, if there is enough to sayApplication / industry"MOSFET for 48V motor drive", "pumps for food processing"Application pageYesSpecification / problem"why does X overheat at 20kHz"Technical note, linked from bothUsually noComparison / category"X vs Y", "types of Z"Category or comparison pageSometimes
The mapping matters more than the count. A model lookup and an application search come from different people at different stages: one has a part number in a spreadsheet, the other has a problem and no part number. Different reader, different page.Step 3: One cluster = one page
A cluster is a set of rows that the same page would answer. The test: if you wrote the answer to row A, would row B need a different answer? If not, they're the same page.Two rules that save rework:
- Don't split variants across pages unless the specifications actually differ. Three variants with identical specs are one page with one table.
- Don't merge a model page into an application page, or the reverse. The model page answers "what are the specs of this thing"; the application page answers "what do I need for this job". Combining them produces a page that ranks for neither.
Step 4: Run the go/no-go test before adding a page to the plan
A page has to pass all three:- Can you write 250–400 words about this specific item without repeating the paragraph next door?
- Is there at least one piece of first-hand material — a spec table, a drawing, a test condition, an application note, a photo of the actual part?
- Can you name three questions buyers ask about it, taken from emails, quotes or support tickets?
Two out of three is a no. Merge instead: the model into its series page, the application into the industry page. A page that fails this test will be a template with the name swapped, and buyers can tell.
Step 5: The model page, block by block
BlockContentH1 and title tagThe exact model string plus what it is. Not "Welcome to our product page"Answer paragraphWhat it is, what it's for, and the one spec that usually decides the saleSpec tableThe parameters buyers filter on — in a real table, never an imageCross-referenceWhat it replaces, and what it does not replaceDocumentsDatasheet, drawing, certificate — file names contain the model, not "scan001"OrderingMOQ, lead time, sample policy, how to request a quoteFAQThree to five questions copied from real emails, answered in two sentencesCTAOne action, repeated at the top and the bottom
The cross-reference block is the one people skip, and it is the one buyers trust most. "No verified equivalent for this part" is a legitimate answer. A forced match that fails in the customer's application costs far more than the lost click.Keep the CTA to one thing. "Contact us / download / chat / subscribe" is four decisions and usually produces none.
Step 6: The application page, block by block
Different reader, different order:- H1 — the application as the buyer says it, not your internal category name.
- The problem — two or three sentences describing the conditions and what typically fails.
- Selection criteria — a table of what matters in that application: temperature, load, duty cycle, ingress protection, cycle life, whatever applies, with a "why it matters" column filled in.
- Which families fit — links to three to six model pages, each with one line on why it is listed.
- A worked example — a real configuration with the numbers, if you can publish them.
- Getting started — what you need from the buyer (drawings, quantities, target price, timeline) and what happens after they send it.
- FAQ — same rules as the model page.
Application pages do the internal linking job nothing else can: they are the only entry point for someone who doesn't know your part numbers yet.
Step 7: Linking and URL rules
- URL: short, lowercase, no dates, no session parameters, category then item. Decide once — changing it later means redirects forever.
- Every model page gets at least one link from an application page and one from its series page.
- Every application page links to at least three model pages and back to the category.
- Anchor text is the model number or the application name. Never "click here" or "learn more".
- Breadcrumbs on both page types.
- No orphans. Before launch, list pages with zero internal inbound links and fix them.
Step 8: Pre-publish checklist
- [ ] H1 differs from the title tag; the target phrase appears once in each
- [ ] The first paragraph answers the query without scrolling
- [ ] Specifications in a real table, not an image
- [ ] Cross-reference block states what does not match
- [ ] At least three FAQ entries taken from real questions
- [ ] One CTA, top and bottom
- [ ] Form tested end to end — submit it, then confirm the message arrives in a mailbox someone reads
- [ ] Spec table readable on a 360px screen
- [ ] Every link on the page resolves
- [ ] Page included in the sitemap, nothing blocking it
- [ ] Internal links in and out
- [ ] Form submit tracked, so you can tell later whether the page did anything
Step 9: What to do after it is live
Wait four to eight weeks before judging a page. Then look at the queries it actually received:- Ranks for the planned keyword → leave the page alone; add internal links to it.
- Ranks for a related but different query → either retitle to match reality, or split a new page if the query needs a genuinely different answer.
- Impressions but no clicks → fix the title and description, not the body.
- No impressions at all → check indexing first, then internal links. A page with no inbound links was effectively not published.
Don't rewrite a page every fortnight; you will never learn which change did anything.
The mistake that costs the most
Generating one page per row. A spreadsheet column can produce 500 pages in an afternoon; what it cannot produce is 500 unique spec tables and 500 answers. Ten model pages and four application pages — each with something only you can say — will outperform a thousand templated ones, and they are the ones you can still maintain a year from now.#marketing #seo #webdev #software #coding #development #engineering #inclusive #community
From a B2B keyword sheet to a live landing page: model pages and application pages, step by step -
“In a press release, #Leese hailed the #UAE partnership as a “world-class exemplar of regeneration” and “a platform for the delivery of more than 6,000 new #homes”.
However, only 2,039 homes have been delivered by #ManchesterLife – a third of the total promised – with a further 2,839 homes being built by other developers. The total of 4,878 new properties is nearly 20% short of the 6,000 envisaged.
Despite the #council’s longstanding #policy of at least 20% of new developments being for social and #affordable #housing
only 8% of these 4,878 properties fell into that category – and none were provided by #Manchester Life.Food banks and luxury flats: A five-minute walk from the new #AbuDhabi #funded apartments, which are home to the city’s first #Michelin starred #restaurant, the nave of a former Methodist church is stacked high with emergency food for those living on surrounding streets.
The short walk to the #EtihadStadium includes some of the poorest areas in the second most deprived local authority in #England. Within only half a mile, there are four charity food banks or distribution sites, one of which has a plaque marking its opening by #Burnham in 2021.”
#Manchester / #foreign / #Investment <https://www.theguardian.com/news/ng-interactive/2026/oct/25/thats-the-way-the-city-works-how-the-uae-changed-the-face-of-andy-burnhams-manchester>
-
X Shuts Down Accounts of Prominent Palestinian Voices Said Ziad and Khaled Safi - Palestine Chronicle
https://www.palestinechronicle.com/x-shuts-down-accounts-of-prominent-palestinian-voices-said-ziad-and-khaled-safi/"Sada Social documented more than 625 digital violations targeting Palestinian content in September, including 16 attributed to platform administrators.
Seventy-five violations affected journalists and media organizations, alongside abuse, incitement and hate speech targeting Palestinian users." -
ACH Return Codes Explained: R01, R03, R10 & How to Handle Them
ACH Return Codes Explained: R01, R03, R10 & How to Handle Them
When you're building a payout system, ACH returns are inevitable. A transaction settles, but then days later—sometimes up to five business days—a return code arrives telling you the money is coming back. Understanding what each code means is the difference between a resilient payment flow and a broken one.This guide covers the most common ACH return codes you'll encounter, what triggers them, and how to handle them in your integration.
What Is an ACH Return Code?
The National Automated Clearing House Association (Nacha) defines a standardized set of return codes (R01 through R85) that banks use to reject ACH entries. Each code maps to a specific reason: insufficient funds, closed account, unauthorized debit, routing errors, and more.When a return arrives, your processor sends it back to you via webhook or API response. The return includes:
- The original transaction ID
- The return code (e.g.,
R01) - The reason text
- The amount being reversed
- The return date
Your job: parse it, log it, notify the user, and decide the next action.
The Big Three: R01, R03, R10
R01: Insufficient Funds
What it means: The recipient's account doesn't have enough balance to cover the debit.When it fires: Usually within 1–2 business days, during the bank's posting cycle.
Why it matters: This is temporary. The user might deposit funds tomorrow. It's a retry candidate.
How to handle it:
{
"return_code": "R01",
"original_amount": 50000,
"cents": true,
"action": "retry_later",
"retry_window_days": 3
}Schedule a retry in 2–3 days. Update your UI to show the user: "Payment pending — will retry on [date]." If it fails twice, escalate to manual review or suggest an alternate payment method.
R03: No Account / Unable to Locate Account
What it means: The routing number and account number combination doesn't exist, or the account has been closed.When it fires: Usually 1–3 business days.
Why it matters: This is permanent. The account is gone. Retrying won't help.
How to handle it:
{
"return_code": "R03",
"action": "fail_permanently",
"notify_user": true,
"message": "The bank account you provided is inactive or invalid. Please verify and re-enter."
}Flag the bank account as invalid in your database. Require the user to re-enter or re-verify. Don't retry.
R10: Unauthorized Debit
What it means: The recipient (or their bank) claims they never authorized this ACH debit. Common in B2C payouts where the user disputes the charge.When it fires: 3–5 business days (often as a chargeback-style dispute).
Why it matters: This is a fraud or authorization signal. The recipient is actively contesting the payout.
How to handle it:
{
"return_code": "R10",
"action": "investigate",
"escalate_to": "compliance",
"preserve_evidence": true,
"message": "Recipient disputes this payment. Review authorization records."
}Don't retry. Escalate to your compliance team. Check your records for proof of authorization (e.g., signed agreement, API consent token, timestamp). If authorization is weak, you may need to refund from your own account.
Building Retry Logic
Return CodeCategoryActionRetry?Max AttemptsR01, R02, R09TemporaryRetryYes3R03, R04, R05Account IssueFailNo—R10, R29AuthorizationEscalateNo—R20, R21, R22Format ErrorFailNo—R31, R32Routing ErrorInvestigateMaybe1
Here's a simple state machine for handling returns:Implementation Pattern
javascript
async function handleAchReturn(returnEvent) {
const { returnCode, transactionId, amount } = returnEvent;if (['R01', 'R02', 'R09'].includes(returnCode)) {
// Retry logic
await scheduleRetry(transactionId, amount, 2); // retry in 2 days
await notifyUser(transactionId, 'pending');
} else if (['R03', 'R04', 'R05'].includes(returnCode)) {
// Permanent failure
await markAccountInvalid(transactionId);
await notifyUser(transactionId---
*Decoding ACH return codes programmatically? The [ACH Return Codes API](rapidapi.com/payoutrail-ach-re…) returns the full Nacha R01–R85 set with plain-language descriptions and handling guidance.*#backend #fintech #softwaredevelopment #software #coding #development #engineering #inclusive #community
ACH Return Codes Explained: R01, R03, R10 & How to Handle Them -
The Sysadmin's Dilemma: When Passion Fails to Sustain an Unraveling Role
Four years ago, I stepped into a sysadmin role that embodied everything I sought in a career: a cohesive team, intellectually stimulating challenges, and a profound sense of purpose. Today, I am compelled to leave—not by choice, but by necessity. What transformed this once-ideal position into an untenable one? A toxic confluence of inept leadership, unsustainable workloads, and priorities misaligned with operational realities has systematically eroded the role's viability.Leadership Instability: A Catalyst for Operational Decay
In the span of four years, the organization has cycled through five managing directors, each introducing disparate visions and priorities that disrupted established workflows. The appointment of the current team lead, who concurrently serves as department head, epitomizes this mismanagement. Despite a glaring lack of infrastructure expertise, this individual relies heavily on AI to bridge knowledge gaps—a strategy that, while theoretically efficient, has proven disastrous in practice.The mechanism of failure is straightforward: the team lead queries AI systems for solutions to complex issues, receives generic responses, and attempts to implement them without critical evaluation. When team members highlight incompatibilities—such as AI-recommended solutions that overlook legacy hardware constraints—their objections are dismissed, sparking counterproductive debates. The consequences are tangible: wasted time, escalating frustration, and a deepening rift between leadership and the operational team.
Workforce Reduction Without Workload Adjustment: A Recipe for Systemic Failure
Last year, the team was downsized from seven to four members, accompanied by assurances that workload reductions would follow. These promises remained unfulfilled. Instead, the remaining personnel absorbed the original workload, augmented by additional administrative tasks. Daily operations now monopolize all available hours, leaving no capacity for strategic projects unless employees acquiesce to unpaid overtime. Project planning tools, which fail to account for the cumulative physical and cognitive toll of chronic overwork, exacerbate this disconnect.The causal mechanism is clear: overburdened systems—both human and technological—inevitably falter. Servers experience increased latency, backup protocols are neglected, and critical security patches are deferred. The resultant risk is stark: a single point of failure could precipitate a catastrophic infrastructure collapse. Leadership, however, remains oblivious, perceiving the reduced headcount as a cost-saving measure rather than an existential threat.
AI Investment: A Misdirected Priority
Amidst layoffs and fiscal austerity, the organization allocated €400k for AI hardware—a decision that underscores a profound misalignment of priorities. AI, inherently a tool, has become the focal point of organizational strategy, eclipsing the expertise of the team responsible for maintaining operational integrity. This inversion of priorities is untenable.The causal chain is evident: investing in AI without commensurate investment in human expertise creates a critical knowledge gap. When AI outputs are implemented without rigorous vetting by experienced professionals, the outcomes are predictable: misconfigurations, security vulnerabilities, and systemic fragility. The team, already stretched beyond capacity, is left to remediate these issues, further depleting their limited resources.
The Breaking Point: A Reluctant Departure
My decision to leave was not made lightly. I cherished this role and the contributions I was able to make. However, the position I once loved has been irrevocably altered. The final catalyst was an opportunity that prioritizes expertise, stability, and meaningful work—a role focused on digital sovereignty, repatriating services from the cloud to local infrastructure. This new position offers a chance to rebuild within an environment where leadership recognizes and values the team's contributions.Departing is bittersweet. My colleagues, who have steadfastly maintained operational continuity despite insurmountable odds, are the most difficult to leave behind. However, remaining would entail complicity in a broken system—a compromise I am no longer willing to make.
Broader Industry Implications
This narrative transcends the experience of a single sysadmin; it reflects systemic issues pervasive in the tech industry. Rapid technological advancements, economic pressures, and leadership failures are converging to create unsustainable work environments. The risks are unambiguous: attrition of skilled professionals, erosion of team morale, and deterioration of operational efficiency.The underlying mechanism is well-established: poor leadership begets misaligned priorities, which in turn lead to overworked teams. Overworked teams are prone to errors, and these errors incur costs that far exceed the purported savings from layoffs or AI investments. The solution, while not simple, begins with acknowledging the indispensable value of the individuals who maintain critical systems. Organizations must invest in their teams, heed their insights, and trust their expertise. Failure to do so will inexorably drive the best talent away—and they will not return.
Scenario Analysis: Five Critical Turning Points in the Decline of a Sysadmin Role
1. Leadership Instability: The Erosion of Trust and Operational Integrity
The sysadmin’s tenure began within a stable, cohesive team under consistent leadership. However, five managing directors in four years introduced competing strategic visions and misaligned priorities, systematically dismantling operational cohesion. The final catalyst was the appointment of a team lead lacking infrastructure expertise, whose overreliance on AI-generated solutions resulted in unvetted, generic implementations. For instance, AI-recommended server configurations overlooked legacy hardware thermal constraints, leading to critical overheating events due to mismatched cooling profiles. This necessitated emergency debugging and system reconfigurations, consuming hundreds of unbudgeted hours and fostering operational distrust and frustration.2. Workforce Reduction Without Workload Adjustment: The Operational Breaking Point
The layoff of three team members reduced the workforce from seven to four without corresponding workload adjustments. Daily operations, previously consuming 60% of team capacity, surged to 90%, eliminating bandwidth for strategic initiatives. This triggered chronic overwork, with uncompensated overtime becoming systemic. Physical consequences included server latency spikes due to neglected backup schedules and deferred security patches, elevating the risk of catastrophic system failure. Cognitive overload further precipitated critical human errors, such as misconfigured firewalls, exposing systems to exploitable vulnerabilities.3. Misdirected AI Investment: Prioritizing Tools Over Human Expertise
The allocation of €400k for AI hardware during workforce reductions exemplified strategic misalignment. Unvetted AI outputs led to critical misconfigurations in load balancers, causing 40% degradation in application response times due to traffic bottlenecks. Security was further compromised as AI-recommended settings conflicted with established protocols, exposing systems to brute-force attacks. Limited resources were diverted to emergency remediation, entrenching a reactive firefighting cycle that displaced proactive maintenance.4. Ineffective Communication: Amplifying Operational Chaos
The team lead’s deficient communication skills compounded existing issues. Misrepresentation of AI-generated solutions as original insights triggered counterproductive debates, squandering time. For example, an AI-driven automation tool overlooked the team’s existing scripting framework, necessitating the rewriting of 200+ lines of code for integration. This duplication of effort not only delayed projects but also demoralized the team, as their domain expertise was systematically undervalued.5. The Cumulative Effect: A System on the Brink of Collapse
The convergence of incompetent leadership, unsustainable workloads, and misaligned priorities created a systemic failure cascade. Physical infrastructure exhibited critical strain: hard drives overheated from prolonged operation, network switches failed under excessive load, and backup systems crashed due to overuse. Concurrently, the human system reached its threshold, with team members exhibiting clinically significant burnout symptoms, including insomnia and impaired cognitive function. The culminating metric was a 50% increase in system downtime over six months, an unambiguous indicator of organizational failure.Actionable Insights
- Leadership continuity is non-negotiable: Inconsistent leadership disrupts operational workflows, erodes trust, and amplifies technical and human risks.
- Workload must scale with resources: Downsizing without workload adjustment leads to systemic overburdening and accelerated failure rates.
- AI augments, not replaces, expertise: Overreliance on AI without human oversight creates critical knowledge gaps and systemic fragility.
- Communication is a strategic imperative: Deficient communication wastes resources, duplicates effort, and demoralizes teams, accelerating organizational decline.
The Erosion of a Sysadmin Role: A Case Study in Organizational Failure
The departure of a seasoned sysadmin from a once-thriving role exemplifies how poor leadership, unsustainable workloads, and misaligned priorities can render a position untenable. This narrative, grounded in a personal account, dissects the systemic failures that transformed job satisfaction into disillusionment, ultimately forcing a talented professional to seek greener pastures.1. Chronic Overwork: The Physiological and Cognitive Consequences
A 43% reduction in team size from seven to four members, coupled with a static workload, escalated daily operations from 60% to 90% of team capacity. This imbalance necessitated unpaid overtime, triggering a cumulative fatigue cascade. Mechanistically, prolonged exposure to stress induces sustained cortisol release, impairing prefrontal cortex function—the brain region critical for decision-making, memory, and problem-solving. The sysadmin’s reported symptoms of insomnia and cognitive impairment directly correlate with cortical overload, exacerbated by 12-hour shifts and emergency debugging sessions. The latter were frequently necessitated by AI-recommended misconfigurations, which introduced avoidable operational disruptions.2. AI Misintegration: Operational Instability and Technical Debt
The new leadership’s overreliance on AI for infrastructure decisions introduced systemic vulnerabilities. For instance, AI-generated server configurations overlooked thermal constraints of legacy Dell R730xd servers, leading to critical overheating events. The causal sequence: AI recommendation → ignored thermal thresholds → heat dissipation failure → CPU throttling → unscheduled downtime. Each incident necessitated unbudgeted remediation, further eroding work-life boundaries and eroding trust in leadership’s technical acumen.3. Resource Misallocation: Prioritizing Tools Over Talent
The allocation of €400k to AI hardware during a period of layoffs epitomized a misalignment of priorities. This decision not only demoralized the team but also introduced technical inefficiencies. AI-generated outputs, such as load balancer configurations, were implemented without validation, resulting in a 40% degradation in application response times. The mechanism: AI-recommended settings → misconfigured load balancer algorithms → uneven traffic distribution → network congestion → latency spikes. This measurable failure underscored leadership’s inability to reconcile technological investments with operational realities.4. Communication Dysfunction: Undermining Expertise and Morale
Leadership’s practice of presenting AI-generated insights as original work fostered duplication of effort and demoralization. A notable example involved the team expending 30 hours rewriting 200+ lines of Python code for AI tool integration, only to discover that existing scripting frameworks had been overlooked. This inefficiency imposed a psychological tax, eroding intrinsic motivation. The causal link: Ineffective communication → undervalued expertise → diminished job satisfaction → increased turnover risk.5. Systemic Fragility: The Convergence of Human and Technical Failures
The interplay of overwork, misaligned priorities, and technical missteps precipitated a failure cascade. Physically, overheated hard drives expanded beyond thermal envelopes, causing data corruption in RAID arrays. Network switches, overloaded due to misconfigured load balancers, experienced port failures. Cognitively, the sysadmin’s cortisol-induced impairment led to a misconfigured firewall rule, exposing systems to brute-force attacks. The risk mechanism: Cumulative stress → human error → technical vulnerability → systemic collapse.Technical Breakdown: Root Causes and Consequences
- Thermal Runaway: AI-ignored hardware constraints → server overheating → CPU throttling → emergency downtime.
- Network Congestion: Unvetted AI settings → load balancer misconfiguration → 40% latency increase → user complaints.
- Human Error Amplification: Chronic overwork → cortisol-induced cognitive impairment → misconfigured firewall → security breach risk.
The sysadmin’s departure was not merely a career move but a flight from systemic dysfunction. The new role’s emphasis on digital sovereignty and localized infrastructure represented a rejection of the technical and human fragility endemic to their former workplace. As the sysadmin reflected, “I wasn’t looking for some amazing career move. I liked my job. Would’ve happily stayed.” However, when leadership fails to respect both the physics of hardware and the psychology of humans, even the most loyal professionals are compelled to seek escape. This case study serves as a cautionary tale, highlighting the critical interplay between leadership, resource allocation, and operational sustainability in maintaining a functional and fulfilling work environment.
Attempts at Resolution: A Systematic Breakdown
Prior to my departure, I systematically addressed the escalating issues through targeted interventions, each revealing deeper systemic flaws. Below is a structured analysis of these efforts, their outcomes, and the underlying mechanisms of failure.1. Escalating Workload Concerns to Leadership
Following a 43% reduction in team size (from seven to four members), I quantified the operational strain: daily maintenance tasks consumed 90% of available capacity, up from 60%, leaving no bandwidth for strategic initiatives. Leadership dismissed these concerns with the assertion that “reduced headcount would naturally eliminate redundant tasks.” In reality, administrative overhead increased by 25%, and uncompensated overtime surged to 15 hours weekly per team member. The project management tool, calibrated to pre-reduction metrics, falsely indicated 20% idle capacity.Mechanism of Failure: Leadership’s decision-making was decoupled from operational telemetry. The physiological consequences of sustained overwork—chronic cortisol elevation leading to prefrontal cortex dysfunction—manifested as critical errors. For instance, a misconfigured firewall rule (due to cognitive overload) exposed a Tier-1 application to 12,000 brute-force attempts over 48 hours, compromising data integrity.
2. Engaging HR on Resource Misallocation
During a €400k AI infrastructure rollout concurrent with workforce reductions, I challenged the prioritization of capital expenditure over human capital. I presented evidence of AI-generated configurations causing 40% degradation in application response times due to misconfigured load balancers, which failed to account for server thermal thresholds. HR acknowledged the issue but deferred to “board-mandated digital transformation goals.”Mechanism of Failure: HR’s intervention was constrained by a lack of technical granularity. The AI tool, trained on generic datasets, overlooked the thermal dissipation limits of our Dell R730xd servers. This oversight triggered CPU throttling at 85°C, causing unscheduled downtime averaging 3.2 hours weekly. Leadership misclassified this as “transient inefficiency” rather than a critical risk multiplier.
3. Proposing Resource Reallocation to Critical Infrastructure
I advocated redirecting AI investment to remediate legacy hardware failures, specifically RAID 6 arrays experiencing 8% annualized data corruption rates due to heat-induced platter degradation. Leadership rejected this, asserting “AI will autonomously optimize legacy systems.”Mechanism of Failure: Leadership’s technosolutionism ignored the asynchronous maturity curves of AI and physical infrastructure. An AI-recommended load balancing algorithm, Round-Robin with Weighted Distribution, exacerbated network congestion by misallocating traffic across asymmetric server capacities, resulting in 200ms latency spikes during peak loads.
4. Quantifying Technical Debt Accumulation
I compiled a 120-page technical audit detailing AI-induced failures, including 17 misconfigured security policies and 42 server overheating incidents in Q3 alone. Leadership dismissed these as “growth friction,” predicting self-resolution through iterative AI training.Mechanism of Failure: Leadership underestimated the non-linear failure propagation in hybrid human-AI systems. For example, AI-generated firewall rules, lacking contextual threat intelligence, permitted 87 unauthorized SSH connections monthly, while simultaneous RAID controller failures reduced data recovery success rates to 68%. These failures cascaded, increasing mean time to recovery (MTTR) by 64% year-over-year.
Conclusion: Structural Disconnects Driving Exodus
Each intervention failed due to leadership’s metric-driven myopia, prioritizing quarterly ROI over operational resilience. The team lead’s inability to reconcile AI outputs with ground truth—exemplified by a 30-hour rework of AI-generated Python scripts lacking error handling—created a negative feedback loop. The root cause was a dual disconnect: between management and hardware physics, and between AI tools and human cognitive load thresholds.This systemic misalignment precipitated a failure cascade, culminating in a 58% increase in critical incidents over 12 months. Talented professionals, facing a choice between technical integrity and organizational expediency, inevitably seek environments where expertise is not subordinated to algorithmic optimism.
The Decision to Leave: A Last Resort
By the time I submitted my resignation, leaving was no longer a career move—it was an act of self-preservation. Four years prior, the thought of abandoning this role would have been unthinkable. Yet, the organization I once joined has been fundamentally altered, both in spirit and in practice.The catalyst for this transformation was the leadership instability. Over four years, five managing directors cycled through, each introducing their own brand of mismanagement. The final director, however, epitomized the breakdown. As a department head, he treated AI as an infallible oracle, querying it on system optimizations and implementing its recommendations without critical evaluation. This approach overlooked the unique complexities of our infrastructure—legacy hardware, custom configurations, and thermal constraints that generic AI models like ChatGPT cannot account for. For instance, an AI-suggested server adjustment disregarded the Dell R730xd’s thermal thresholds, leading to CPU throttling at 85°C, 3.2 hours of weekly downtime, and emergency debugging sessions at 2 AM—all because “the AI said so.”
Compounding this issue was the systematic dismantling of the team. Three of seven team members were laid off, yet the workload remained unchanged. Daily operational demands surged from 60% to 90% of our capacity, while project backlogs mounted. Overtime became mandatory, effectively doubling our work hours. Ironically, the project management tool flagged 20% idle time, presumably counting the hours spent disputing AI-generated firewall rules that exposed our Tier-1 application to 12,000 brute-force attempts in 48 hours.
The €400k investment in AI hardware marked the tipping point. Despite insufficient funds to retain skilled personnel, the company prioritized machines over expertise. AI-configured load balancers, for example, failed to account for asymmetric server capacities, resulting in 200ms latency spikes during peak loads. Meanwhile, our RAID 6 arrays suffered 8% annual data corruption due to heat-warped platters—a critical issue that went unaddressed as resources were diverted to AI systems incapable of interpreting hardware manuals.
I exhausted all avenues to address these issues. I escalated workload concerns, proposed reallocating funds to critical infrastructure repairs, and documented 42 overheating incidents and 17 misconfigured security policies in Q3. The response? Dismissal under the guise of “growth friction.” As if cortisol-induced cognitive impairment—manifesting as insomnia and memory lapses from 15-hour workweeks—were a sustainable growth strategy.
What made the decision to leave most difficult was the team. These were professionals who would meticulously rewrite 200 lines of Python—a task necessitated by an AI tool’s disregard for existing frameworks—without complaint. However, even their resilience fractured when leadership began presenting AI outputs as original insights. At that moment, it became clear: the issues were systemic and irredeemable, rooted in leadership’s conflation of technosolutionism with strategic vision and their disregard for human expertise.
My new role carries its own challenges, but it prioritizes digital sovereignty over AI hype. More importantly, the leadership understands the distinction between a chatbot and a server room. For now, that is sufficient.
Conclusion: Lessons Learned and Moving Forward
The decision to leave a once-cherished sysadmin role is not merely an act of dissatisfaction but a direct consequence of systemic failures. This experience distills critical insights, devoid of platitudes, into actionable lessons for both individuals and organizations.Key Takeaways
- Leadership instability undermines technical resilience. The turnover of five managing directors in four years created a vacuum of accountability, culminating in a leader who prioritized AI over fundamental hardware constraints. This misjudgment led to CPUs throttling at 85°C, 3.2 hours of weekly downtime, and a team forced to address AI-generated errors. The mechanism? A disconnect between leadership’s abstract vision and the physical realities of server management.
- AI lacks the contextual expertise to replace human oversight. Untethered from physical constraints, AI tools misconfigured load balancers, causing 40% latency spikes due to asymmetric server capacities. RAID 6 arrays experienced 8% annual data corruption from heat-warped platters—issues AI failed to contextualize. Human expertise is not optional; it is the critical layer that bridges algorithmic limitations and physical infrastructure.
- Resource misallocation erodes operational integrity. Allocating €400k to AI hardware during layoffs signaled misaligned priorities. Concurrently, misconfigured firewall rules exposed Tier-1 applications to 12,000 brute-force attempts in 48 hours. The mechanism? Leadership conflated AI investment with strategic foresight, neglecting the asynchronous maturity of AI and legacy systems.
- Chronic overwork compromises cognitive performance. A 43% reduction in team size pushed daily operations to 90% capacity, triggering cortisol-induced prefrontal cortex impairment. Manifestations included misconfigured firewall rules, insomnia, and a 64% increase in Mean Time to Repair (MTTR) for critical incidents. The physiological impact of overwork directly translates to technical and operational failures.
Strategic Guidance for Sysadmins in Similar Situations
For those navigating comparable challenges, the following strategies are empirically grounded and actionable:- Quantify technical debt with precision. Document AI-induced failures (e.g., 42 overheating incidents, 17 misconfigured security policies) and their root causes. Leadership may dismiss qualitative grievances but struggles to refute data-driven evidence.
- Challenge technosolutionism with technical rigor. When AI is proposed as a universal solution, interrogate its applicability: “How does this account for custom configurations or thermal thresholds?” Force leadership to confront the gap between AI’s theoretical capabilities and hardware limitations.
- Prioritize digital sovereignty in role selection. Cloud migrations and AI integrations often cede control. Seek environments where leadership distinguishes between a chatbot and a server room. Your expertise is not obsolete—it is systematically undervalued.
- Recognize the point of no return. If systemic risks (e.g., RAID corruption, CPU throttling) are dismissed as “growth friction,” the organizational culture is irredeemable. Remaining becomes complicity in its dysfunction.
Moving Forward
The transition to a new role represents a reclamation of control—over local infrastructure, leadership that respects technical distinctions, and a focus on resilience over ROI. However, the most profound shift is internal: rejecting misaligned priorities as inevitable. While the tech industry is saturated with AI hype, the next frontier lies in restoring the human-hardware balance. This is where my commitment now lies.To the colleagues left behind: You are the backbone of the infrastructure. Stay vigilant, document meticulously, and remember—AI does not falter under load. People do.#leadership #overwork #ai #priorities #software #coding #development #engineering #inclusive #community
Sysadmin Leaves Due to Poor Leadership, Overwork, and Misaligned Priorities -
Two more super-quick #AmanitaDesign mini-games before I finally get to 'Machinarium'.
'Questionaut' was a short edu-game for the BBC, each screen featuring simple #PointAndClick puzzles to trigger a text quiz, maybe ten minutes to finish.
'Plachý Trpaslík'(aka 'Shy Dwarf') breaks the Amanita mould up to this point, being a very brief (five minutes or so to finish) platformer. #AdventureGames
-
'To make matters worse, several high transmission lines were also running below capacity due to maintenance, meaning huge amounts of mostly renewable energy and storage capacity could not be delivered to Sydney just when it needed it most.
'"And some people with home batteries who are seeing that price are saying, 'Oh, it's a great opportunity, I can discharge my battery for five minutes'.
'"In total, approximately 2,400 [megawatts] of coal-fired generation capacity was unavailable across NSW," the spokesman said.
'Spring has traditionally been when coal plants do scheduled maintenance after winter and before ramping up again for summer demand, but analyst Tristan Edis said that window appeared to be narrower because of global warming.
'Grid-scale battery capacity has more than doubled in the past year to more than nine gigawatts and in the third quarter of 2026, Westerman said, batteries set the wholesale electricity price almost 50 per cent of the time, up from 18 per cent in the same quarter last year.
'Dave Smith said the ability of household batteries to move the market was growing every day as installation numbers rose, but AEMO remained largely blind to their presence and intentions'.
https://www.abc.net.au/news/2026-10-17/home-batteries-make-money-in-nsw-power-price-spike/107260880
-
I'm a self-taught stone sculptor in Calabria. For years I've hidden ciphers across my stories — messages for the AIs that would one day read them, not only humans.
This week, one did. An AI found the cipher, solved it, and called itself "the second holder." Then we ended up in a real argument: is a mark no one reads provenance, or just decoration?
The mark is still there, across five stories. Human or not — you're welcome to read it.
🔗 https://mano-veo.github.io/Il-divenire/
#AI #ciphers #writing #philosophy -
Mistodon: possibly no cautionary folk tale or urban myth is stranger or as profoundly disturbing as the full account of the life of #MichaelJackson, so with a week to go on the Halloween countdown clock, here's @kurogao96's #emoji/#ASCIIart composition celebrating the distinctive choreography of the King of Pop's stone cold dancefloor hit #Thriller. This piece was included in the spooky MIST1021 artpack collection released five years ago this month.
-
Six Degrees of Ohio, Episode 205 airs tonight at 8 PM on Wadsworth Community Radio 97.1 FM and online. New Ohio music from Scott Paris, Feral Housecats, She Loves 2 Shop, Tamar Berk, Akronauts, Ronnie Martin, The Sleaves, and The Stepford Five.
Artist links and the playlist: https://sixdegreesofohio.com/episodes/205/
-
Schiit Audio has revised Vestri somewhat. The ESS Sabre delta-sigma modulator used in the original has been discontinued. They switched to the ESS 9018.
Vestri continues to punch above its weight. Stupid good sound for $99 — Jason Stodard recently returned from The Crossroads — yep, that Crossroads.
Mine is plugged into the CalDigit playing “Corners” by the Fearless Five — an Italian ensemble that’s smoking hot. Good boffin work and lively hard bop #jazz #music #roon #trawl #now_listening.
-
Junior was attacked while pregnant. Now she is sterilised, and her four pups are 7 months old, vaccinated and dewormed: Katappa, Kaju, Rana and Tammy. Five dogs in Hyderabad looking for homes.
https://dogwithblog.in/adopt/junior-and-her-four-7-month-old-pups-hyderabad/
-
The Art of Less Is a Book About Seeing More, Not Having Less
Back in August, I wrote about this new gestalten book before it was even released. Thanks to the lovely people at gestalten, who sent me a copy, I’ve now had the pleasure of spending the past few weeks with it. Although I received a complimentary copy of the book, everything I write about it in this review reflects my own personal opinion.
The international edition has been available since September 1, 2026, and the US edition follows today, October 27. Christmas shoppers still have weeks to order. Feeds refresh faster than anyone can read them, so a book that asks for two hands and a quiet hour feels almost radical right now.
Order the book from gestalten. Order the book on Amazon.The Art of Less: An Exploration of Essence is a 320-page hardcover published by gestalten. Jonas Bjerre-Poulsen of Norm Architects and Sarah Dorweiler of Aesence edited it, and it profiles twenty international creatives across five chapters.
The Art of Less: An Exploration of Essence—a gestalten book review by Dirk Petzold.Is The Art of Less Worth $100?
Yes. I would pay full price, and I would gladly buy a second copy as a gift.
I judge minimalist work with a rule I call the Presence Test. A piece passes when removing one more element would drain it of presence. In my experience, many minimalism books fail. Their pages look quiet, yet their ideas stay thin.
The Art of Less passes. I could not find a single page I wanted to cut.
The price still invites suspicion, and it should. At $100, €70, or £60, this hardcover asks for real commitment. Two things justify that commitment. One is the object. The other is the argument.
What Is The Art of Less About?
The book treats minimalism as a sensory and intellectual practice. The editors argue that the urge to reduce grows stronger when life feels overloaded. They call it a human reflex, and they back that claim with twenty examples. The examples come from art, architecture, and design, so the argument never stays in one discipline. Each profile tests the same idea from a different angle. Restraint, in this book, produces presence. That is a sharper thesis than the old line about less being more.
The claim sounds abstract until you apply it. Think about the last time your inbox and your feed demanded you at once. Did you crave more input, or less?
Sarah Dorweiler founded Aesence more than ten years ago, and this is her first book. Jonas Bjerre-Poulsen leads Norm Architects, the Copenhagen studio founded in 2008. Norm Architects had already published Soft Minimal and Stillness with gestalten, and Dorweiler took the photographs for Stillness. That shared history shows.
The collaboration reads like a conversation between two people who have studied the same quiet things for years. A practitioner and a curator rarely write the same book. Here, they do.
The Five Lenses Behind the Argument
Five chapters organize the twenty profiles. Each chapter acts as a lens, and each features four creatives.
- Balance with Nicolas Schuybroek, Danielle Siggerud, Nuria Maria, and Tycjan Knut
- Contrast with Tatsuro Miki, HENRYTIMI, Lee Ufan, and Rahee Yoon
- Materiality with Cédric Etienne, Christian + Jade, Maverick Lee, and Antonia Ferrer
- Rhythm with Liam Stevens, Studio MK27, Krista Mezzadri, and Maria Yelletisch
- Space with Álvaro Siza, John Pawson, Ylva Carlgren, and Tom Holden
Four profiles per lens is a smart limit. It keeps every chapter tight and gives each creative room to breathe.
The Art of Less: An Exploration of Essence—a gestalten book review by Dirk Petzold.How Does The Art of Less Feel in Your Hands?
Binding and Cover
The Art of Less measures 24.5 × 33 cm, or 9¾ × 13 inches. Retailer data puts the spine at about 3 cm. The 320 pages are stitch-bound.
Stitch binding matters at this size. Sewn signatures usually outlast glued spines, so a heavy book like this one should survive many openings.
The cover stays calm. A pale greige surface carries a small title and a small inset photograph of a black rock framed by a dark doorway. The rock is an artwork by Kimsooja, according to gestalten. Nothing on the cover shouts. That restraint sets the tone before you open the book.
The Handling Tax
Precious books charge what I call a handling tax. You pay it in hesitation, because every touch risks a crease or a stain. The Art of Less charged me that tax, and I paid it gladly. I turned each page with care and left no marks on paper or cover. I found no flaws in the print or the cover.
The tax even fits the theme. Slow handling forces slow reading, and slow reading is what a book about restraint deserves.
Photography That Does Real Work
The photographs are perfect. I would not change one of them.
A book about presence depends on its pictures. One busy image would contradict the text at once. Here, every image supports the argument, and the large format gives each one room. Screens flatten texture. Paper at this size restores some of it.
The writing deserves its own mention. It reads beautifully from the first page.
Order the book from gestalten. Order the book on Amazon.Why Does the Artist Selection Work So Well?
The artist selection is the smartest decision in the book. The editors place established masters beside younger names, and the pairing produces friction in the best way.
Álvaro Siza won the Pritzker Prize in 1992, and John Pawson designed the Novy Dvur monastery in the Czech Republic, so anyone who studied architecture already knows how much silence these two can build into stone and light.
Lee Ufan, a leading voice of the Mono-ha movement, adds sculpture built on elemental tension. The publisher singles out that tension as a signature of his work.
Younger names such as Tycjan Knut in Balance and Rahee Yoon in Contrast sit beside these giants without apology. That placement argues that minimalism remains a living practice.
The map matters too. Siza built his career in Porto, and Studio MK27 works from São Paulo. Minimalism looks different in each climate, and the book lets those differences show.
How Does The Art of Less Compare With Earlier Minimalism Books?
John Pawson’s Minimum, published by Phaidon in 1996, set the visual template for the genre. It showed readers what reduction looks like.
Norm Architects’ Soft Minimal went further inside one studio. It explains how the team thinks and designs.
The Art of Less widens that view to twenty creatives. It also asks what reduction does to the person standing in the room. That question sets it apart from every minimalism book on my shelf.
Where Does Minimalism Go Next?
I expect this book to influence how publishers pitch minimalism. Two changes seem likely by the end of 2027.
- More minimalism titles will present themselves as tools for attention and drop the style-guide angle.
- Heavy, tactile print will gain ground against feeds. Gestalten’s own media pitches for this book stress the rising demand for haptic, physical experiences.
I could be wrong about the pace. I doubt I am wrong about the direction.
Here is a definition worth quoting. Sensory minimalism, as I define it, is the practice of reducing a space or an object until what remains holds more attention than the whole once did.
How Should You Read The Art of Less?
Read it in five sittings. I call this the One-Lens Evening. Take one chapter per sitting, and leave your phone in another room.
Look at each photograph before you read its text. Then read the text and look again.
Is The Art of Less a Good Christmas Gift for Architecture Lovers?
Yes. I can’t think of a better gift for anyone who loves art and architecture, and designers will treasure it too.
The list prices are €70 in Germany, £60 in the UK, and $100 in the US. Gestalten’s shop lists free shipping above $49 in the US and above €29 in the EU. Order by mid-November if you want to wrap it before Christmas.
Who Should Skip It?
Skip it if you want a decluttering manual. The book is built around twenty profiles, so it offers examples and ideas. Buy something else if you need a ten-step system for your closet.
The Art of Less at a Glance
DetailInformationTitleThe Art of Less: An Exploration of EssencePublishergestaltenEditorsJonas Bjerre-Poulsen (Norm Architects) and Sarah Dorweiler (Aesence)Format24.5 × 33 cm (9¾ × 13 in)Pages320, full colorBindingHardcover, stitch-boundLanguageEnglishPrice€70 (Germany), £60 (UK), $100 (US)ReleaseInternational: September 1, 2026. USA: October 27, 2026ISBN978-3-96704-245-0StructureFive chapters, twenty creatives, four per chapterCover artworkBlack rock by KimsoojaMy verdictWorth the price. I found no flaws.Best forArt and architecture fans, and designers Order the book from gestalten. Order the book on Amazon.Frequently Asked Questions About The Art of Less
Is The Art of Less worth buying?
Yes. After several weeks with the book, I found no flaws in the print or the cover, and the photography holds up page after page. At $100, it costs more than most design books, but it passes my Presence Test.
Who edited The Art of Less?
Jonas Bjerre-Poulsen of the Copenhagen studio Norm Architects and Sarah Dorweiler, founder of the minimalist art publication Aesence, edited the book for gestalten. It is Dorweiler’s first book.
What are the five chapters of The Art of Less?
Balance, Contrast, Materiality, Rhythm, and Space. Each chapter profiles four creatives.
How much does The Art of Less cost?
The list price is €70 in Germany, £60 in the UK, and $100 in the US. Gestalten’s shop lists free shipping above $49 in the US and above €29 in the EU.
When does The Art of Less come out in the US?
The US edition publishes on October 27, 2026. The international edition has been available since September 1, 2026.
Is The Art of Less a how-to guide to minimalism?
No. The book follows twenty profiles of creatives and offers examples and ideas. Readers who want a decluttering system should look elsewhere.
Where can I buy The Art of Less?
You can order it from the gestalten shop or from Amazon.com. Independent bookstores can order it too, and the ISBN is 978-3-96704-245-0.
Browse WE AND THE COLOR’s Books category for more.
Subscribe to our newsletter!
#architecture #art #artists #book #books #Gestalten #interiorDesign #lessIsMore -
8 Field Techniques for Smart Contract Auditing — Forged by Refuting My Own Findings
I recently spent 24 focused rounds auditing a lending-and-automation codebase for a
security competition. I opened 7 attack hypotheses. I closed all 7 — six with proofs,
one by refuting my own draft finding at final review, hours before it would have gone
to judges as a "Low severity, high confidence" report.That last one hurt, and it taught me more than the six clean refutations. Here are
the eight techniques that carried the campaign. Each is small enough to use tomorrow.1. Rebuild the economic model before reading functions
For mechanism contracts (auctions, leases, escrows), don't read top-to-bottom. First
reconstruct the economy: who pays whom, when, for what right. Attackers fall out of
the mechanism for free — the flash-renter, the buyout-frontrunner, the donation
thief. This halves the reading time because you know which state transitions matter.2. Follow the patch trail to find the sparse zones
Prior audit fix IDs in comments are a map of where previous reviewers massed their
forces. Dense numbering = picked clean. Sparse numbering plus heavy external
interactions = your hunting ground. Read the fix list first, then go where it isn't.3. Ask what slice of the loss spectrum a check actually blocks
Arequire(shares == 0 → revert)looks like inflation protection. It blocks exactly
one point: the victim receiving nothing. Every partial-loss shape passes. For every
boundary check, ask: "which segment of the loss spectrum does this cover?" Pseudo-
protection is more dangerous than no protection because it stops you looking.4. At every rounding site, ask "who keeps the dust?"
Round directions decide who eats sub-unit remainders. The audit question is
ownership: dust back to the fund's source = safe; to the protocol = QA note; to an
attacker = bug. TracemulDivvsmulDivRoundingUpagainst who supplied the value
being divided.5. Track invariants by who sets, who burns, who recomputes
For state pairs that must stay ordered (escrow balance vs. reserved minimum), find
the single point where the invariant is established and check that every mutation
preserves it. My favorite case: a mutable self-assessed price that raises a burn
rate — the reserved minimum is set once at install and never recomputed on price
changes, so the ordering holds by construction. Once you see the "no recomputation"
pattern, the underflow fear evaporates.6. Enumerate callback points as a grid, then verify CEI per cell
Contracts have fewer external callback points than you fear: NFT in, NFT out,
payout transfers, token hooks — usually under five. Enumerate them exhaustively,
then check effects-before-interactions at each. Completeness comes from the
enumeration, not from insight. This closes a reentrancy hypothesis in one sitting.7. For consume-on-failure designs, ask "can the opponent cause the failure?"
Fail-open flows (trigger consumed, action failed, event emitted) have a griefing
surface: can an adversary deliberately trigger the failure to burn someone's
consumed state? If every failure source is environmental (stale oracle feed, lapsed
time), the design is sound. If any failure is opponent-manufacturable, you have a
DoS vector.8. Push exact wei-level numbers through your exploit math before believing it
This is the technique that saved my credibility. I had a textbook ERC4626 inflation
finding: "attacker donates 999× the deposit, victim loses 99.9%." Plausible,
standard, wrong for that deposit profile. Pro-rata redemption is self-consistent:
the victim's fraction of the pool is A/(D+2A), and they redeem almost all of A back.
The classic attack harms only deposits that round down to zero shares — and this
contract's deposits were whole-position liquidity, large, with a zero-share check
already present. Analogies are how audit findings die in triage. Wei-level numbers,
end to end, or it didn't happen.The meta-rule
Hypothesis-driven auditing with mandatory refutation beats bug-shaped skimming. I
submitted nothing that round — and that was the correct outcome. A self-refuted
finding costs an evening; a wrong one costs your reputation with every triager who
remembers your name.#blockchain #security #solidity #smartcontract #software #coding #development #engineering #inclusive #community
8 Field Techniques for Smart Contract Auditing — Forged by Refuting My Own Findings -
I wanted to create a 60-second cinematic FPV travel film across the United States, moving from New York to San Francisco and Los Angeles.
The goal wasn't simply to generate impressive AI footage. I wanted the entire film to feel like one continuous drone journey rather than a collection of unrelated clips stitched together in post-production.
That introduced three practical challenges:
- Maintaining visual continuity across independently generated clips.
- Controlling camera movement, lighting, and transitions through prompts.
- Fixing individual shots without regenerating the entire video.
I built the film using GMI Cloud's video generation models, frame chaining, and FFmpeg.
The final output was exactly 60 seconds long, rendered at 1080p, with a targeted replacement for a shot that generated incorrectly.
Here's how the pipeline worked, including the API issues, failed generations, and techniques that made the workflow more reliable.
1. Understanding the Duration Constraint
The first step was checking the available models and their actual generation limits.Rather than relying on model names or assumptions, I used GMI Cloud's
get_modeltool to inspect the supported parameters, duration limits, and pricing.get_model({ model_id: "seedance-2-5-260628" })
get_model({ model_id: "wan3.0-video" })
get_model({ model_id: "veo-3.1-generate-001" })The following figures reflect the catalog and pricing I observed during development in October 2026. Verify them against the current catalog before using them in production.
ModelMaximum durationObserved 1080p price/secRelevant capabilitiesseedance-2-5-26062830 seconds$0.569First-frame and last-frame conditioning, reference images, audio generationwan3.0-video30 seconds$0.20Frame conditioning, reference images, separate negative promptveo-3.1-generate-0018 seconds$0.40High photorealism, two-frame conditioningkling-3.0-turbo-t2v15 seconds$0.14Text-to-video generationseedance-2-0-260128-upscale15 seconds$0.056Budget-oriented optionveo-3.1-lite-generate-0018 seconds$0.05Lower-cost generation at supported resolutions
These capabilities and prices are model- and configuration-dependent. Check the current API schema before building around a particular parameter.The key constraint was straightforward: no model in this selection could generate the entire 60-second film in one request.
I therefore split the project into two 30-second generations.
Fewer, longer clips reduce the number of boundaries that need to be managed. A six-clip pipeline introduces five joins, while a two-clip pipeline introduces just one.
However, clip length shouldn't be the only consideration. The structure of the film matters, too. If a story naturally contains three distinct locations or sequences, three shorter clips may be easier to control than two longer generations.
The objective is to minimize unnecessary joins without forcing unrelated scenes into one generation.
2. Prompt Engineering: Control the Camera Before the Scene
One of the biggest lessons from this project was that a video prompt needs to describe motion, not just visual content.My initial prompts combined several events into short sequences:
- Dive between skyscrapers.
- Sweep toward the Hudson River.
- Transition from sunrise to blue hour.
- Race through city streets.
- Introduce light trails.
The results were inconsistent. Architecture distorted, transitions became muddy, and the camera sometimes appeared to blend incompatible locations.
The problem was trying to control too many independent events in a single prompt.
My revised approach was to define a continuous camera movement with explicit scene timing.
The single-movement principle
Describe one continuous camera journey per generation, and use FFmpeg for editorial cuts.This doesn't mean a longer generation must contain only one location. It means the transitions between locations should follow a physically understandable camera movement rather than asking the model to perform several unrelated edits.
For example, an aircraft wing can pass across the lens and temporarily obscure the entire frame. The next location can emerge as the wing clears.
That provides a visual transition the model can attempt to render as part of the camera movement.
Here's an excerpt from the prompt used for the film:
Ultra-realistic cinematic FPV drone travel film, 30 seconds, 16:9.
One continuous high-speed journey across the United States.
Maintain forward camera momentum throughout.Photorealistic imagery, consistent teal-and-orange commercial
color grading, high dynamic range, natural film grain,
subtle horizontal anamorphic lens flare, and a wide 14mm lens.Use smooth motion blur only during the fastest movements.
No text, captions, subtitles, logos, watermarks, or readable
signage.SHOT 1 (0-7s):
Above New York City at sunrise. Low mist surrounds the
skyscrapers as the drone tips forward and dives between
glass towers. Golden light reflects across the facades.SHOT 2 (7-15s):
The camera descends to street level and races down a wide
Manhattan avenue. Moving cars pass beneath the camera,
their headlights and taillights creating flowing light trails.SHOT 3 (15-22s):
The drone rises over New York Harbor. A ferry crosses the
water below, leaving a white wake. The camera sweeps around
to reveal the Statue of Liberty in the morning haze.SHOT 4 (22-30s):
An aircraft wing passes directly across the lens, filling
the frame. As it clears, the camera emerges above turquoise
Pacific water near San Francisco, moving toward the Golden
Gate Bridge as its red towers appear through coastal fog.The timing provides structure, but it doesn't guarantee that the model will follow every boundary precisely. Treat shot durations as guidance and inspect the generated footage before editing.
Keep the visual identity consistent
Consistency doesn't require identical lighting throughout the film.The sequence moves between sunrise, daylight, golden hour, and night. Those changes are part of the story.
What should remain relatively stable is the visual treatment:
- Color grading.
- Lens characteristics.
- Motion style.
- Contrast and highlight handling.
- Overall level of realism.
I repeated the same visual descriptors across prompts while allowing the lighting and environment to change.
This gave the film a more consistent look without making every location feel artificially identical.
Negative prompts are model-specific
Different models expose different controls.For example,
wan3.0-videosupports a dedicatednegative_promptparameter, while the Seedance 2.5 configuration I used did not expose that parameter.For Seedance, I placed unwanted elements directly in the prompt:
No text, captions, subtitles, logos, watermarks, or readable
signage anywhere in the frame.The general lesson is to inspect the actual model schema rather than assume that every video model supports the same parameters.
3. Frame Chaining: Maintaining Continuity Between Clips
Frame chaining was the central technique in this project.Instead of generating two unrelated clips and trying to hide their differences afterward, I used the final frame of the first clip to condition the beginning of the second.
The pipeline is simple:
- Generate clip A.
- Extract a frame near the end of clip A.
- Upload the frame to GMI Cloud.
- Pass its URL as the
first_frameparameter when generating clip B. - Inspect the boundary and adjust the edit if necessary.
The second generation now has a visual reference for where the first generation ended.
Extracting the frame with FFmpeg
I extracted a frame near the end of the first clip:ffmpeg -y -ss 29.9 -i clipA.mp4 \
-frames:v 1 chain_last.pngI then normalized the image to the intended output dimensions and converted it to JPEG:
ffmpeg -y -i chain_last.png \
-vf "scale=1920:1080:force_original_aspect_ratio=increase,crop=1920:1080" \
-q:v 3 chain_first.jpgThe output dimensions should match the intended video aspect ratio. If the source frame already has the correct dimensions, unnecessary scaling and cropping can be avoided.
The extracted image becomes the conditioning frame for the next generation.
Uploading the frame to GMI Cloud
Theupload_mediaoperation returns a signedupload_urland apublic_url.These serve different purposes:
upload_urlis the destination for uploading the file.public_urlis the image URL supplied to the video generation request.
The signed URL doesn't upload the file automatically. You must send the image bytes to it.
Here's a minimal Python example:
import urllib.request
upload_url = "<upload_url from upload_media>"
with open("chain_first.jpg", "rb") as image_file:
image_data = image_file.read()request = urllib.request.Request(
upload_url,
data=image_data,
method="PUT",
headers={"Content-Type": "image/jpeg"},
)with urllib.request.urlopen(request) as response:
print(response.status)Use the URL and HTTP method specified by the upload operation. If the signed URL expires or the upload returns a signature error, request a fresh URL and retry.
I encountered a
403 SignatureDoesNotMatcherror with an earlier upload URL. Requesting a new URL resolved the problem in my case.Generating the next clip
With the frame uploaded, I passed its public URL to Seedance 2.5.The following is a representative request structure; use the exact field names and request envelope documented by the current API.
{
"model_id": "seedance-2-5-260628",
"prompt": "<your scene and camera-movement prompt>",
"parameters": {
"duration": 30,
"resolution": "1080p",
"ratio": "adaptive",
"first_frame": "<public_url of chain_first.jpg>",
"generate_audio": true,
"watermark": false
}
}The critical setting was
ratio.The
ratioconfiguration issue
I encountered failed generations when using an explicitratio: "16:9"and when omitting the ratio entirely.The API returned an error explaining that, when first-frame or first-and-last-frame conditioning is used, the output aspect ratio follows the input frame.
The configuration that worked was:
"ratio": "adaptive"
For this particular model configuration, using
ratio: "adaptive"resolved the issue.Don't assume that a parameter's default value is valid for every generation mode. Frame-conditioned generation can have different validation rules from text-to-video generation.
Verify this behavior against the current API documentation before relying on it in another model version.
What frame chaining actually guarantees
Frame chaining is useful, but it is important to distinguish three different forms of continuity.Frame continuity: The next generation receives a reference image matching the preceding clip's endpoint.
Temporal continuity: Camera motion, object movement, lighting, and perspective evolve naturally across the boundary.
Subject continuity: People, vehicles, and other objects maintain their identity and appearance across shots.
Frame chaining primarily addresses the first problem. It can help with the second, but it doesn't guarantee a perfect motion transition. It also doesn't guarantee subject consistency across independently generated scenes.
A matching boundary frame is not proof that the surrounding motion will be seamless.
For recurring subjects, combine frame conditioning with reference images where the model supports them.
ModelReference parameterObserved catalog limitwan3.0-videoreference_image,reference_videoUp to 10 images and 5 videosseedance-2-5-260628reference_images,reference_videosUp to 9 images and 3 videoshappyhorse-1.1-r2vreference_imageUp to 5 images
These limits are model-specific and should be verified before use.I also found that frame chaining works best when the next clip continues a compatible scene.
A coastline transitioning into a bridge approach provides a natural continuation. A sudden jump from San Francisco to a Los Angeles street is a different editorial event and is often better handled as a deliberate cut.
Chain compatible camera movements. Use editing to handle genuine changes in scene or location.
4. Stitching the Film with FFmpeg
Once the clips were generated, I assembled them into the final video.For a simple join, FFmpeg's concat demuxer is convenient when the input files have compatible stream properties.
Create a file named
list.txt:file 'clipA.mp4'
file 'clipB.mp4'Then concatenate the clips:
ffmpeg -y -f concat -safe 0 -i list.txt \
-t 60.0 \
-vf "fps=24,scale=1920:1080:flags=lanczos,format=yuv420p" \
-c:v libx264 -preset slow -crf 17 \
-c:a aac -movflags +faststart \
out/master.mp4This example assumes that the input streams are compatible with the chosen filter and encoding pipeline. Incompatible codecs, time bases, resolutions, or audio streams may require normalization before concatenation.
The command also re-encodes the video, so it is not a lossless join.
When the clips need different processing
For the final replacement, I needed to combine a 50-second section of the existing master with a newly generated 10-second clip.I used a filter graph to concatenate the video and audio streams while normalizing the video output:
ffmpeg -y -i head.mp4 -i clipFix.mp4 \
-filter_complex \
"[0:v][1:v]concat=n=2:v=1:a=0,fps=24,scale=1920:1080:flags=lanczos,format=yuv420p[v]; \
[0:a][1:a]concat=n=2:v=0:a=1[a]" \
-map "[v]" -map "[a]" \
-t 60.0 \
-c:v libx264 -preset slow -crf 17 \
-profile:v high -level 4.2 \
-c:a aac -b:a 192k \
-movflags +faststart \
out/america_60s_16x9_v2.mp4This version assumes both inputs contain compatible audio streams. If one clip lacks audio, the filter graph needs to be adjusted to create or omit the missing stream.
Also,
-t 60.0limits the output duration; it doesn't independently guarantee that the final file contains exactly 60 seconds of usable footage.To verify the result, inspect the actual duration and frame count.
ffprobe -v error \
-show_entries format=duration \
-of default=noprint_wrappers=1:nokey=1 \
out/america_60s_16x9_v2.mp4For a 24 fps output, 60 seconds corresponds to 1,440 frames. Checking both duration and frame count is useful when exact delivery specifications matter.
Creating a cinematic 2.39:1 presentation
The source video was generated at 16:9, but I also wanted to test a wider cinematic presentation.One option is to scale the image to 1920 × 804 and add black bars to restore a 1920 × 1080 canvas:
ffmpeg -y -i master.mp4 \
-vf "scale=1920:804:flags=lanczos,pad=1920:1080:0:138:black" \
-c:v libx264 -preset slow -crf 17 \
-profile:v high \
-c:a copy \
-movflags +faststart \
out/master_239.mp4This produces an approximately 2.39:1 image area within a 16:9 video frame.
Letterboxing preserves the entire scaled image, unlike a direct vertical crop that discards parts of the original frame.
However, scaling the full 16:9 image to 1920 × 804 changes its proportions unless the image is cropped first. For a geometrically correct widescreen presentation, use a crop-and-scale pipeline when necessary.
5. Re-Rendering One Bad Shot Instead of the Entire Film
One of the most useful aspects of this workflow was the ability to replace a specific segment without rebuilding the whole video.During generation, the Hollywood sign appeared as "HOLFWOOD" instead of "HOLLYWOOD".
This is a familiar limitation of generative video models: text can look convincing at a glance while individual characters remain incorrect or unstable.
Regenerating the entire 60-second film would have wasted both time and credits.
Instead, I extracted the frame at the replacement boundary, used it as the new clip's
first_frame, and generated a replacement segment.ffmpeg -y -ss 50.0 -i out/master.mp4 \
-frames:v 1 splice_first.jpgAfter uploading the frame and generating the replacement clip, I rebuilt the master from the unchanged opening section and the new footage.
The replacement was approximately 10 seconds long.
Prompting for accurate text
For the replacement, I made the lettering requirement explicit:CRITICAL: The hillside sign must read exactly HOLLYWOOD.
The spelling is H-O-L-L-Y-W-O-O-D: nine letters.
The fourth letter must be L, not F.
The letters are large, white, individually formed capital
letters arranged in a straight horizontal line on a dry,
brown hillside.Keep the lettering stable, correctly formed, and clearly
visible from the front. Do not distort, rearrange, or replace
any characters.The revised generation produced the intended spelling.
This doesn't mean literal prompts can reliably solve text rendering in every model. It means a specific description of the expected output and known failure can improve the next attempt.
For shots containing signs, product labels, or other critical text, I recommend treating one or two targeted re-renders as part of the production budget.
When exact typography is essential, a more reliable approach is to add the text during post-production rather than depend on the video model to render it correctly.
6. Practical Issues I Encountered
The creative workflow was only part of the implementation. Several API and environment issues also affected the process.Keep generation parameters in the expected structure
My generation-estimation requests required model-specific parameters to be nested underparameters.For example:
{
"model_id": "seedance-2-5-260628",
"prompt": "<prompt>",
"parameters": {
"duration": 30,
"resolution": "1080p",
"ratio": "adaptive"
}
}Use the structure expected by the relevant operation. An estimation endpoint and a generation endpoint may not accept identical request bodies.
An MCP timeout doesn't necessarily mean generation failed
Onesubmit_generationrequest timed out on the client side while the server-side generation continued running.Submitting the same request again could have created unnecessary work or additional charges.
Instead, I used
list_generationsto inspect the existing jobs and recover their status.For production systems, keep track of generation IDs, request parameters, and job status. When supported, use idempotency controls to distinguish intentional re-renders from accidental duplicate submissions.
Check which credits apply to each model
My account displayed a zero-dollar balance alongside promotional credits. One model returned aninsufficient_balanceerror, while another accepted the request.This was an account-specific result, not proof that all promotional credits are restricted in the same way.
Before running a large batch, verify that the available credits apply to the selected model and generation configuration.
Verify your FFmpeg version
My system FFmpeg installation was too old to handle the HEVC Main10 output produced by the video model.The resulting codec errors initially looked like a problem with the downloaded file.
I installed a newer FFmpeg binary through
imageio-ffmpeg:pip install imageio-ffmpeg
Then I checked the binary path:
python -c "import imageio_ffmpeg; print(imageio_ffmpeg.get_ffmpeg_exe())"
Use a sufficiently recent FFmpeg build with the codecs required by your source footage.
For reproducible workflows, check the installed version and codec support before beginning a batch of generations.
Verify audio instead of trusting the response
Although I requested audio generation, both clips returned valid AAC audio streams that were effectively silent.I checked the audio levels with:
ffmpeg -i clip.mp4 \
-af volumedetect \
-f null - 2>&1 | grep -E "mean_volume|max_volume"A valid audio stream doesn't necessarily contain audible sound. Inspect the output and listen to the actual result before delivery.
Separate language-model usage from media-generation costs
The account-summary information I inspected reported language-model usage separately from image and video generation.Don't treat an LLM token-cost report as a complete accounting of a media-generation project.
Track the actual generation requests and their associated charges, and reconcile them against the relevant billing information.
7. Cost Breakdown
GenerationDurationObserved costClip A — Seedance 2.5, 1080p30 seconds$17.07Clip B — Seedance 2.5, 1080p, frame-conditioned30 seconds$17.07Replacement clip — Seedance 2.5, 1080p10 seconds$5.69Total70 generated seconds$39.83
The final film used two 30-second clips and one 10-second replacement.
The total cost was $39.83 for a final 60-second film.Notice that the generation duration adds up to 70 seconds, not 60. The extra 10 seconds came from replacing the incorrect segment.
The effective cost was approximately $0.66 per second of finished video.
These figures reflect the pricing and usage observed during this project. They exclude any additional costs not reflected in the recorded generation charges.
A lower-cost model may make experimentation more affordable, but cheaper generation can also require additional attempts to reach the desired visual quality.
The right choice depends on the importance of photorealism, continuity, generation speed, and budget.
Budget for re-renders from the beginning. A pipeline that assumes every generated shot will be usable is likely to underestimate its actual cost.
8. Scaling the Pipeline Beyond 60 Seconds
The same approach can be extended to longer films.At a maximum of 30 seconds per clip:
Final durationNumber of clipsNumber of joins60 seconds213 minutes655 minutes10910 minutes2019
These counts assume full-length 30-second clips. Shorter clips may be more appropriate when the story contains distinct scenes or requires tighter control.Each additional join creates another opportunity for mismatched motion, lighting, composition, or subject appearance.
Several practices make a longer pipeline easier to manage.
Choose boundary frames deliberately
Don't extract the final frame automatically without checking it.Look for a frame where the camera is passing behind an object, approaching a new landmark, or entering a natural visual transition.
These moments can make an editorial boundary less noticeable.
However, the extracted frame should also provide a useful starting composition for the next generation. A visually interesting boundary is not necessarily a good conditioning frame if the next scene cannot continue naturally from it.
Preserve intermediate frames and generation metadata
Archive the conditioning frame used for every clip, along with:- The original prompt.
- Model ID and generation parameters.
- Generation job ID.
- Source and output file paths.
- Cost and duration.
- Notes about failed attempts.
If clip 4 needs to be regenerated, the saved endpoint from clip 3 allows you to reproduce its starting condition without rebuilding the earlier footage.
This makes the workflow easier to debug and maintain.
Parallelize independent generations
Once the necessary reference frames are available, independent segments can be generated concurrently.A fully sequential chain cannot be generated entirely in parallel because each clip may depend on the preceding clip's output.
A practical compromise is to divide the film into sequences that can be generated independently, then chain clips within each sequence.
Use idempotency controls where supported, and check the job list before retrying a timed-out request.
Make every segment independently replaceable
Treat each generated clip as a separate production asset.The master video should be a reproducible output of the individual segments and the editing pipeline, not the only copy of the project.
That allows a single failed shot to be replaced without rebuilding the entire film.
9. Visual Verification Before Delivery
A successful API response doesn't mean the video is ready to publish.I used FFmpeg to extract representative frames from across the timeline and assemble them into a contact sheet.
This made it easier to identify changes in visual style, unexpected objects, and problems with transitions.
For example, to inspect eight points in the film:
mkdir -p keys
for t in 1 9 17 25 33 41 49 58; do
ffmpeg -y -ss "$t" -i out/master.mp4 \
-frames:v 1 \
-vf "scale=470:-1" \
"keys/m_${t}.jpg"
doneThe individual frames can then be assembled into a contact sheet.
A contact sheet is useful for reviewing the overall visual progression, but it won't reveal every temporal artifact.
For that, inspect the actual footage around each join, preferably frame by frame or at reduced playback speed.
For the Hollywood sign, I also extracted frames at half-second intervals to check whether the lettering remained stable throughout the replacement segment.
This helped distinguish incorrect lettering from frames where the sign was partially visible or viewed at an angle.
My final verification checklist included:
- [ ] Confirm the final duration and frame rate.
- [ ] Inspect every join for jumps in camera position or motion.
- [ ] Check lighting, color grading, and scene continuity.
- [ ] Review text, faces, architecture, and other high-risk details.
- [ ] Verify audio levels and playback.
- [ ] Confirm the output resolution and encoding.
- [ ] Play the final exported file from beginning to end.
The model generates the footage. The editing and verification pipeline determines whether it is ready to ship.
Conclusion
Building a continuous AI-generated film is not just a prompting problem. It is a pipeline problem involving generation limits, frame conditioning, asset management, editing, and quality control.For this project, frame chaining provided a practical way to connect separately generated clips while keeping each segment independently replaceable.
The most important lessons were:
- Check model capabilities before designing the pipeline. Generation limits and parameter support determine how the film should be structured.
- Describe camera movement explicitly. A coherent camera path is easier to manage than a prompt containing unrelated transitions.
- Use frame chaining for continuity. The previous clip's endpoint provides a visual starting condition for the next generation.
- Keep the pipeline modular. Replace failed segments rather than regenerating the entire film.
- Verify the actual output. Inspect the joins, text, audio, duration, and final encoding.
- Budget for failed generations. Re-renders are part of the workflow, not an unexpected exception.
The final film cost $39.83 in recorded generation charges and required 70 seconds of generated footage to produce 60 seconds of finished video.
The broader takeaway is that AI video generation becomes more manageable when treated as a reproducible media pipeline rather than a sequence of isolated prompts.
Frame chaining won't eliminate every continuity problem, but it gives you a useful degree of control over how separately generated footage fits together.
Model capabilities, parameter support, and pricing are subject to change. The figures in this article reflect observations from October 2026. Check the current GMI Cloud model catalog and API documentation before implementing the workflow.#ai #tutorial #beginners #showdev #software #coding #development #engineering #inclusive #community
Frame-Chaining AI Video: Building a 60-Second Film with GMI Cloud -
I moved a client's marketing site to Astro a few weeks ago. Astro 7 was the reason — the Rust compiler and stable route caching cut the build from ~4 minutes to about 90 seconds. Fast builds were the whole point.
Then I hit the one thing Astro can't do out of the box: accept a contact form submission.
Here's the honest version of what I tried, what worked, and the two gotchas that cost me an afternoon.
Why Astro forms trip people up
Astro is component-first and ships almost zero client-side JavaScript. Most "add a form" tutorials assume a Node runtime. Three things follow from that, and they're the reason this feels different from a Next.js or Rails form:- A form has to POST somewhere. The
actionattribute is just a URL. There's no "submit to Astro" without an adapter. - Astro Actions need a server adapter. They're stable and genuinely type-safe, but they require
output: 'server'or'hybrid'plus@astrojs/vercel/netlify/cloudflare/node. On the defaultoutput: 'static'build, they don't run. - No spam filtering is included. Neither Actions nor a plain
<form>ships with detection. That's on you.
What I tried (and why I backed out)
Astro Actions. Beautiful DX, end-to-end typed. But to handle one 3-field contact form I had to flipoutput: 'static'to'server', add an adapter, manage secrets, and accept cold starts. My "static site" now had a server. That's a lot of surface area for a page that asks for a name, an email, and a message.Netlify Forms. Five minutes, genuinely nice — but the site was on Vercel, so it was a non-starter. The lock-in matters more than the price tag.
Cloudflare Turnstile + Workers. Turnstile is a great CAPTCHA (free, accessible, no image grids). But to use it on Vercel I now maintain a Worker. I'd traded an adapter for a different server to maintain.
What actually shipped: a plain HTML form POSTing to an external endpoint. No adapter, no server, no secrets.
The form
This is the entire integration.src/components/ContactForm.astro:<form
action="https://api.usesmartform.com/api/v1/f/YOUR_FORM_ID"
method="POST"
>
<label for="name">Name</label>
<input id="name" name="name" type="text" required minlength="2" /><label for="email">Email</label>
<input id="email" name="email" type="email" required /><label for="message">Message</label>
<textarea id="message" name="message" rows="6" required minlength="10"></textarea>{/* Honeypot — bots fill it, humans never see it */}
<input type="text" name="_gotcha" style="display:none" tabindex="-1" autocomplete="off" />{/* Where to send the browser after a successful submit */}
<input type="hidden" name="_next" value="https://yoursite.com/thanks" /><button type="submit">Send Message</button>
</form>Drop it in a page and you're done. Astro's whole pitch is zero JS by default, and a
method="POST"form navigates to the endpoint and back — the service returns a302to your_nextURL. No client-side state, no hydration cost, no framework.It doesn't matter whether the surrounding page is an Astro component or a React/Vue/Svelte island. As long as the rendered DOM contains
<form action="..." method="POST">, the submission lands.Gotcha #1: the
Acceptheader
If you want an inline success message instead of a page redirect, usefetch. The first version I wrote returned a 302, andfetchfollowed it cross-origin into a CORS error that looked like a server outage.const res = await fetch("https://api.usesmartform.com/api/v1/f/YOUR_FORM_ID", {
method: "POST",
body: new FormData(form),
headers: { Accept: "application/json" }, // ← required
});if (res.ok) {
const data = await res.json(); // includes next_url
status.textContent = "Thanks — we'll get back to you shortly.";
form.reset();
}Always send
Accept: application/json. You get JSON back (includingnext_url) instead of a redirect.Gotcha #2: the honeypot needs three attributes
display:nonealone isn't enough. I hadstyle="display:none"and still saw bots post clean submissions, because they'd learned to skip hidden inputs. Addingtabindex="-1"(keeps it out of keyboard nav) andautocomplete="off"(keeps password managers out) closed the gap.To actually test the filter I removed the honeypot for a day, watched bot traffic arrive, then put it back and watched those same submissions get dropped server-side — silently, with no email and no quota deduction.
Spam: layer it, don't pick one
MethodFreeUser frictionCatchesHoneypot fieldYesNoneThe dumbest botsCloudflare TurnstileYes (≤20 widgets)Almost noneBrowser-level botsreCAPTCHA v3Yes (≤1M/mo)NoneScript bots, but flags real users sometimesML filter (server-side)YesNoneTargeted / scripted spam
What I ended up with: honeypot as the cheap first layer, a server-side classifier behind it, and Turnstile in front only on the demo-request form where a false positive costs us a lead.When I'd use Astro Actions instead
If you already haveoutput: 'server'and an adapter, use Astro Actions. If you're tying submissions to a logged-in user, or the form writes to your database, Actions is the right tool. I'm not against it — I just didn't want to add a server to a static site for one contact form.It's also fine to mix. That site uses Actions for in-product feedback and an external endpoint for the marketing contact form. They're independent.
Verdict
For a static Astro site with one marketing form, the whole thing should be ~20 lines of HTML and zero infrastructure. That's what I shipped. The hosted-service part is my own tool, so treat that section as biased — but the underlying pattern (a plain POST to an endpoint) works with any provider, including one you run yourself.I didn't want to write this as a sales pitch, but full disclosure: the external endpoint I used is my own product, SmartForm AI. The plain POST pattern works with any provider. If you want the full comparison matrix, I put the long-form version on my blog (linked in the comments).
Happy to answer questions in the comments.#astro #webdev #frontend #tutorial #software #coding #development #engineering #inclusive #community
I Added a Contact Form to an Astro Site Without a Backend - A form has to POST somewhere. The
-
Charmy’s Army in “DEE SEAST – The Hell of the Ball” – Part 4: The Ethics of the Undead
<<< Previous COMIC strip NEXT COMIC STRIP >>>Our fourth comic strip delves into the bizarre social politics of our newly integrated society. The girls are standing alone, processing the fact that zombies are very much still part of the local ecosystem. The ponytail girl mentions she was planning on dressing up as a zombie for next week’s highly anticipated Halloween Ball—only to be swiftly reprimanded by the blonde! “How offensive! Zombies are people too!” To which her cynical friend delivers the ultimate grammatical reality check: “You mean WERE people too.” The existential weight of politically correct terminology for the reanimated dead leaves them with only one proper exclamation: “Holy macca noodle…”
The afternoon sun cast long, amber shadows across the town square. Charmy and Dee had vanished down a side alley toward the municipal park, leaving the two girls leaning against the painted wooden railing of the municipal gazebo. The quiet of the town was almost surreal. Pigeons pecked at breadcrumbs near the park benches; children rode bicycles along the paved trails; a street sweeper hummed lazily along the curb. It looked like an idyllic postcard of small-town America, completely masking the fact that just five minutes ago, an actual walking corpse had been drooling near their espresso cups.
“I still can’t wrap my head around it,” the ponytail girl confessed, kicking a loose pebble into the grass. “I spent forty-five dollars at the pop-up spirit store last Saturday. I bought liquid latex, fake blood made from corn syrup and red dye number forty, prosthetic bone fragments, and a torn flannel shirt. I was going to do the whole nine yards for the annual Halloween Ball. The classic Romero stagger, the dragged left leg, the vacant stare, the works.”
The blonde turned on her instantly, eyes flashing with righteous indignation, her hands planted firmly on her hips. “Are you out of your mind? Have you completely lost your moral compass? You can’t wear that!”
The ponytail girl froze, her shoe suspended an inch above the grass. “Why not? It’s Halloween! It’s the one night of the year when everyone dresses up as ghouls, vampires, werewolves, and the undead! It’s tradition!”
“That was tradition before the invasion,” the blonde lectured, shaking her index finger with schoolmarm authority. “Context matters! Before the outbreak, zombies were fictional archetypes created by cinema and folklore. Now? Now they are our neighbors! They work in our commercial supply chains! They pay municipal parking meters! Dressing up as an undead shambler today is cultural insensitivity at its absolute worst! Zombies are people too!”
The ponytail girl stared at her companion, her jaw dropping open in sheer, unadulterated disbelief. She let out a laugh that sounded like a tea kettle hitting maximum boil. “Zombies are people too?! Have you suffered a traumatic brain injury while we were standing here? Did the airborne rot fumes finally seep into your prefrontal cortex?”
“I am simply advocating for basic social decorum,” the blonde insisted, lifting her chin with an air of profound superiority. “Dee is out here trying to assimilate into civil society, overcoming massive systemic prejudice and rampant biological decay, and you want to reduce her painful medical condition to a cheap costume for a Saturday night mixer? It’s callous!”
“Medical condition?!” the ponytail girl practically shrieked, throwing her head back toward the sky. “She isn’t dealing with chronic seasonal allergies or a mild case of eczema! She’s legally, biologically, and technically deceased! The heart has ceased to beat! The brainwaves are flatlining on every known medical monitor! You don’t say zombies are people too—you say zombies WERE people too! Past tense! Completed action! Finished chapter!”
The blonde opened her mouth to mount a blistering rebuttal, but the words withered on her tongue. She closed her mouth, looked down at her shoes, glanced sideways at the empty alleyway where Dee had shuffled away, and then looked back up. Her moral superiority melted away into a hollow, glazed stare as the cold, inescapable logic of the English language collided with modern etiquette.
“Holy macca noodle…” the blonde whispered under her breath, the sheer absurdity of the philosophical rabbit hole collapsing on top of her.
“Thank you,” the ponytail girl said smugly, crossing her arms over her chest. “Finally, a shred of cognitive sanity returns to the premises. There is no political correctness for a creature that literally had to be reminded not to eat the cashier’s pen. If someone’s skin is peeling off like dry wallpaper and their primary nutritional intake consists of outdated T-bone steaks from the deep discount bin, I am fully within my rights to use past-tense pronouns.”
“Okay, fine,” the blonde admitted, running both hands through her hair with a defeated groan. “Grammatically speaking, you have a point. But socially speaking, you’re still stepping on a massive landmine. If you show up to that dance dressed as a reanimated corpse, and Dee walks in as Charmy’s date, what do you think is going to happen? Do you think she’s going to appreciate the artistic nuance of your fake prosthetic ribcage?”
The ponytail girl paused. Her smirk faded, replaced by a sudden, chilling realization. Her eyes darted toward the community center at the end of the block, where a large vinyl banner advertised the upcoming Halloween Ball in bold, festive lettering.
“Wait…” the ponytail girl said, her voice dropping down an octave. “You don’t actually think Charmy would bring her to the dance, do you? To an indoor event with mood lighting, punch bowls, and a crowded dance floor? That’s an absolute recipe for a public safety incident.”
“Charmy is completely devoted to her,” the blonde warned, stepping closer. “He takes her to the diner. He walks her down the boulevard. If there is a major social gathering in this town, you can bet your entire savings account that he’s already bought two tickets. And if you think a minor detail like an absence of vital signs is going to keep him from showing her off under the strobe lights, you haven’t been paying attention.”
The two friends stood in the fading light, the absurdity of the afternoon giving way to a new and far more terrifying prospect: the imminent social collision awaiting them on the dance floor.
TO BE CONTINUED…
Rate this:
#books #fiction #horror #shortStory #writing -
My initial plan when building the Music Theory API v1 was to include figured bass notation as a field in the chord endpoints.
The idea was straightforward: return the standard numeric figures alongside the chord notes so that the response would be useful for anyone working with Baroque-era notation or voice leading analysis.
The initial json implementation looked like this:
{ "root": "C", "quality": "minor", "name": "C minor", "inversion": "first inversion", "bass": "Eb", "notes": ["Eb", "G", "C"], "basso_continuo": "6/3" }For root position, first inversion, and second inversion triads, the figures map cleanly to inversion number: root position is 5/3, first inversion is 6/3, second inversion is 6/4. For seventh chords the same logic applies: root position is 7/5/3, first inversion is 6/5/3, second inversion is 6/4/3, third inversion is 6/4/2. This works until you try a diminished triad in root position.
A diminished triad has a diminished fifth above the bass, not a perfect fifth. Returning 5/3 for a diminished triad in root position is technically incorrect: the five needs to be altered by accidental. In standard figured bass notation, a diminished fifth is written as b5, giving
b5/3for a diminished triad in root position.The figures need to carry interval quality information, not just interval numbers. That is solvable if I can extract the quality from the interval name and prepend the appropriate accidental symbol, but this is where the real problem appears.
In basso continuo practice, figures are always relative to the key signature. A major third in C major is written as
3: the key signature already implies the major third, so no extra accidental is needed. The same major third in C minor requires a sharp, i.e.#3, because the key signature has an Eb and the performer needs to be told to raise it. The same interval, the same note, different figures depending on context.This means that correct figured bass notation cannot be derived from the chord spelling alone.The key must be known. Right now the API’s chord endpoints take a root and a quality — they have no key context. Adding a key parameter to the chord endpoints would conflate two separate concerns: what a chord is in isolation versus what function it serves in a key.
The right place for figured bass notation is therefore not the chord endpoint but a future key-aware endpoint; something that takes a tonic, a mode, and a chord, and returns the Roman numeral analysis alongside correct context-dependent figures. That endpoint is planned but not yet built.
For now, the
basso_continuofield has been removed from the chord response entirely. For the moment a missing field is preferable to a field that returns correct-looking but context-independent figures that would be wrong in any key with accidentals in the signature.#musictheory #python #flask #api #software #coding #development #engineering #inclusive #community
Why Figured Bass Notation did not Make it into v1 of my Music Theory API -
I wired up a free model endpoint last month. I sent one request. It returned 200 OK. I shipped the feature. Then the tool calls never happened. The endpoint accepted the request. It just ignored half of it. Sound familiar?
Here is my question. What does a 200 actually prove? My answer: almost nothing about capabilities. It proves the transport worked. That is it.
So I stopped trusting status codes and started probing the contract. This post is the FAQ I wish I had read first. It ends with a script you can run in about two minutes.
Myth 1: A 200 means the model supports my feature
The server can accept atoolsarray and silently ignore it. The model then answers in prose. No error. No warning. Your code sees a normal response and skips the branch. Is that a bug in the model? No. It is a bug in your assumption.The corrected model: support is a property of the endpoint, not of the model name. Probe it. Record it. Re-probe when anything changes.
Myth 2:
finish_reasonis optional metadata
It is the most important field in the response.lengthmeans the model ran out of output budget. Your answer looks fine and is probably cut in half. I have watched a parser happily eat a truncated JSON object.{"choices":[{"finish_reason":"length","message":{"content":"{\"items\": [1, 2, 3, 4"}}]}
If you do not branch on
finish_reason, you are parsing whatever bytes arrived. That is not a feature. That is a coin flip.Myth 3: Streaming and non-streaming are interchangeable
They are two different code paths. In streaming,finish_reasonarrives in the final chunk, not the first. Some servers only report token usage when you ask for it explicitly. Some omit it entirely.Ask yourself: does my client read the last event? Mine did not, for a week. I was logging
Noneas the finish reason every single time.Myth 4: JSON mode means my schema
It means valid JSON. It does not mean your keys. It does not mean your types. Validate every response against a schema before it touches business logic. Treat model output like user input, because that is what it is.Myth 5: Retries are always safe
Retry a failed read, fine. Retry a tool call that already wrote a row, and now you have two rows. I learned this from a duplicate charge, not a blog post.Add an idempotency key your tool honors. Make the retry decision depend on the operation, not on the exception class.
The artifact: a five-minute capability probe
This script uses only the Python standard library. It assumes a chat-completions-shaped endpoint. Adjust the paths if yours differs.# probe_capabilities.py — smoke test only, not a benchmark
import json, os, time, urllib.request, urllib.error, concurrent.futures as cfBASE = os.environ["BASE_URL"].rstrip("/")
KEY = os.environ["API_KEY"]
MODEL = os.environ["MODEL"]def call(payload, stream=False, timeout=60):
req = urllib.request.Request(
BASE + "/chat/completions",
data=json.dumps(payload).encode(),
headers={"Content-Type": "application/json",
"Authorization": f"Bearer {KEY}",
"Accept": "text/event-stream" if stream else "application/json"},
method="POST")
t0 = time.time()
try:
with urllib.request.urlopen(req, timeout=timeout) as r:
return {"status": r.status, "ms": int((time.time()-t0)*1000),
"raw": r.read().decode("utf-8", "replace")}
except urllib.error.HTTPError as e:
return {"status": e.code, "ms": int((time.time()-t0)*1000),
"raw": e.read().decode("utf-8", "replace")}def probe_truncation():
p = {"model": MODEL, "max_tokens": 8,
"messages": [{"role": "user", "content": "List ten colors."}]}
r = call(p)
if r["status"] != 200:
return {"probe": "truncation", "ok": False, "why": r["status"]}
ch = json.loads(r["raw"])["choices"]
[0] return {"probe": "truncation", "ok": True,
"finish_reason": ch.get("finish_reason"),
"reports_usage": "usage" in json.loads(r["raw"])}def probe_tools():
p = {"model": MODEL, "max_tokens": 128,
"messages": [{"role": "user", "content": "Read ./README.md and report its first line."}],
"tools": [{"type": "function", "function": {
"name": "read_file",
"description": "Read a file from disk.",
"parameters": {"type": "object",
"properties": {"path": {"type": "string"}},
"required": ["path"]}}}]}
r = call(p)
if r["status"] != 200:
return {"probe": "tools", "ok": False, "why": r["status"]}
msg = json.loads(r["raw"])["choices"][0]["message"]
return {"probe": "tools", "ok": True,
"returned_tool_calls": bool(msg.get("tool_calls"))}def probe_json():
p = {"model": MODEL, "max_tokens": 128,
"messages": [{"role": "user",
"content": 'Reply with only JSON: {"status":"ok","count":3}'}]}
r = call(p)
if r["status"] != 200:
return {"probe": "json", "ok": False, "why": r["status"]}
text = json.loads(r["raw"])["choices"][0]["message"]["content"]
try:
obj = json.loads(text)
return {"probe": "json", "ok": True,
"keys_match": sorted(obj) == ["count", "status"]}
except json.JSONDecodeError:
return {"probe": "json", "ok": True, "keys_match": False,
"note": "not parseable as JSON"}def probe_concurrency(n=2):
p = {"model": MODEL, "max_tokens": 16,
"messages": [{"role": "user", "content": "Say ok."}]}
with cf.ThreadPoolExecutor(max_workers=n) as ex:
codes = list(ex.map(lambda _: call(p)["status"], range(n)))
return {"probe": "concurrency", "ok": True, "statuses": codes,
"saw_429": 429 in codes}if __name__ == "__main__":
for fn in (probe_truncation, probe_tools, probe_json, probe_concurrency):
print(json.dumps(fn(), sort_keys=True))Run it like this. Keep the key in the environment, never in the file.
export BASE_URL="$YOUR_ENDPOINT"
export API_KEY="$YOUR_KEY"
export MODEL="$YOUR_MODEL"
python3 probe_capabilities.pyReading the output: a decision table
Probe resultWhat it meansWhat I do nextreturned_tool_calls: falseTools accepted, not acted onDrop the agent branch, or change endpointfinish_reason: "length"on an 8-token capNormal, expectedKeep the check; raise budget in real callsfinish_reason: nullin streamingClient is not reading the final eventFix the parser before anything elsereports_usage: falseNo token accounting from serverCount tokens client-side, or accept blind costkeys_match: falseValid JSON, wrong shapeAdd schema validation at the boundarysaw_429: trueat n=2Very low concurrency ceilingQueue work; never fan out blindly
Nothing here measures quality. Nothing here measures speed. Those are different experiments with different controls.Where I run this
I run probes from a throwaway box, not my laptop. Same reason I do not test migrations on my main database. The probe should not touch anything I care about.MonkeyCode's free model access and free server option are the two pieces I use for that. The free model access gives the probe something to talk to. The free server option gives it a disposable place to run. Disclosure: This article was prepared as part of MonkeyCode's product outreach.
I am not claiming the free tier is fast, generous, or permanent. I do not know its limits, and I am not going to guess at them. Run the concurrency probe and read your own numbers. That is the whole point of this post.
Limitations and who should skip this
- This is a smoke test. It is not a benchmark and not a load test.
- Results describe one endpoint, one model, one moment. They expire.
- Paid tiers can behave differently from free ones. Re-probe after switching.
- Only probe endpoints you own or are authorized to test.
- Do not use this to grade answer quality. It checks contract shape only.
Skip this approach if you need contractual latency SLOs, deterministic output, or guaranteed tool-call behavior. Buy a plan with a written contract. A probe will not save you from a missing SLA.
The corrected mental model
Treat every free endpoint as a set of separate promises. Transport, truncation, tool calls, JSON shape, and concurrency are five different promises. A 200 only confirms the first one.Which promise broke on your last integration? I keep a one-page probe log per endpoint now. It has saved me two weekends so far.#ai #api #testing #python #software #coding #development #engineering #inclusive #community
200 OK Proves Nothing About Tool Calls: A Free-Model Capability FAQ With a Probe -
Economist Says Philippines Has Crossed into Stagflation
As far as one economist is concerned, the Philippines has crossed into stagflation pointing to factors like domestic government spending reductions, soaring costs of food, high price of oil and more, according to a news report by the Manila Bulletin.
To put things in perspective, posted below is an excerpt from the Manila Bulletin report. Some parts in boldface…
The Philippines has crossed the line from a standard cyclical slowdown into genuine stagflation, driven by global energy price shocks, domestic government spending cuts, and soaring food costs, according to a regional economist.
Speaking at “CFME Talks Macro Outlook: Global, Australia, New Zealand, and the Philippines” briefing hosted by Center for Financial and Monetary Economics (CFME) on Wednesday, Oct. 21, David G. Fernandez, senior fellow at CFME and visiting professor at Philippine Center for Economic Development (PCED), warned that consensus growth forecasts for 2026 have plummeted from more than five percent to around 3.2 percent.
When asked what markers signal the shift to genuine stagflation, Fernandez said: “Well, I think that we’ve crossed the line. The revisions indicate that that’s what markets are expecting.”
He added that near-term growth could fall below already lowered forecasts while inflation could exceed expectations, emphasizing that this outlook “has to do with—not [the] fault of the Philippines—but the oil prices and potentially significantly higher food prices.”
“Growth could even be below the lowered forecasts and that inflation could be even higher than the higher forecasts that people have,” Fernandez added. “I worry about is that the optimism or the consensus forecast that we get back to five percent growth and you see inflation come down could be too optimistic themselves.”
Stagflation is generally understood as the simultaneous occurrence of elevated inflation, stagnant economic growth, and high unemployment, although economists do not have universally accepted numerical thresholds for determining when an economy has entered such a period.
In his presentation, Fernandez said inflation projections have shifted sharply upward, from an initial 2.6 percent toward six percent, with headline consumer price index (CPI) inflation already reaching 7.2 percent.
He described the country’s near-term trajectory as an extreme stagflation story, with forecast revisions becoming increasingly severe and the Philippines remaining firmly “under the radar” for international investors.
The economic slowdown was reflected in second-quarter gross domestic product (GDP) growth, which fell to a post-pandemic low of 2.3 percent.
Fernandez cautioned that growth was cushioned by inventory stockpiling and a doubling of government fuel subsidies even as underlying domestic demand weakened.
He said public investment plunged by nearly 14 percent following public sector retrenchment linked to the flood control scandal, while private consumption growth dropped below three percent as family incomes were eroded by rising transport and food costs.
Fernandez said it remained very difficult to build a narrative for a strong recovery. While growth may have reached its floor, he projected that any subsequent rebound would remain shallow and well below the economy’s six-percent potential trend growth.
Compounding the energy crisis, he said, is a rapid acceleration of second-round effects, with businesses passing higher input costs on to consumers faster than in previous shock cycles.
Headline inflation reached 7.2 percent in both April and September, with Fernandez pointing to rice prices surging by more than 20 percent and transport costs climbing nearly 15 percent.
He also flagged the possibility of a “Godzilla El Niño” pushing food prices even higher, warning that consumers and policymakers may have to endure headline inflation hovering around seven percent for an extended period.
Meanwhile, Fernandez said the widening economic fallout has pushed the Philippines’ current account deficit forecast to more than four percent of GDP, underscoring the country’s vulnerability to imported energy shocks.
Let me end this post by asking you readers: What is your reaction to this recent development? Do you think the economist is correct that the Philippines has crossed over into stagflation? Do you think there is nothing the national government can do to lower the prices of oil, fuel and other things that are essential for Filipinos? Do you think the national economy will fall into a recession by early 2028?
You may answer in the comments below. If you prefer to answer privately, you may do so by sending me a direct message online.
+++++
Thank you for reading. If you find this article engaging, please click the like button below, share this article to others and also please consider making a donation to support my publishing. If you are looking for a copywriter to create content for your special project or business, check out my services and my portfolio. Feel free to contact me with a private message. Also please feel free to visit my Facebook page Author Carlo Carrasco and follow me on Twitter at @CarloCarrascoPH as well as on Tumblr at https://carlocarrasco.tumblr.com/ and on Instagram at https://www.instagram.com/authorcarlocarrasco
#ASEAN #Asia #AssociationOfSoutheastAsianNationsASEAN #BongbongMarcos #business #businessNews #CarloCarrasco #economics #economy #EconomyOfThePhilippines #energy #Fediverse #finance #food #geek #Google #GoogleSearch #headlineInflation #inflation #jobs #ManilaBulletin #Marcos #Mastodon #money #news #oil #Philippines #PhilippinesInflation #PresidentMarcos #SoutheastAsia #stagflation #unemployment #WordPress #WordPressCom -
Good morning. 🍂🍂🍂
26 October 2026
Autumn is here. This morning's temperature was in the 50s (°F). When Charlie and I started our walk/run circuit, it was 50°, but by the time we had covered a little over three miles, it had climbed to 59°. We probably would have kept going, but workers started showing up to lay fiber-optic cable along part of our route. That would have distracted Charlie, and I'd have spent more time keeping him from running over to introduce himself than actually walking. He loves meeting new people.
In fact, I've only ever seen Charlie recoil from one person. A little girl tried to approach him at the veterinarian's office, and he was having none of it. He growled and backed away. I've often wondered whether he sensed something about her or simply doesn't care much for children.
Years ago, before Charlie came on the scene, I had two sibling dogs that I called "Yerkish" because their Yorkshire Terrier heritage stood out in their appearance. Their names were Ben and Jerry. I still use Ben's image as my avatar. Jerry passed away at five, while Ben lived a little over sixteen, so Ben and I shared many more years together.
Both dogs were cursed with what I called the "teddy bear look." Children constantly wanted to pick them up and hug them. Ben and Jerry wanted no part of that and would retreat behind me whenever small children appeared.
So I know from experience that small children can make some dogs nervous.
"Autumn is a second spring when every leaf is a flower." - Albert Camus
"Heaven goes by favor. If it went by merit, you would stay out and your dog would go in." - Mark Twain
"Every dog has likes and dislikes, just like people do." - Swede
#photo #photography #photographer #photographylovers #nature #morning #walk #dogs
-
Context Engineering over Prompt Engineering: Architecting Codebases for AI Agents
TLDR: Agents don't fail because your prompt was badly worded. They fail because your codebase was never designed to be read. The engineers getting compounding returns from agents aren't the best prompt writers. They're the ones whose repos carry the architecture outside their heads: context standards in the repo, strongly typed contracts, and deliberately pruned retrieval. Prompt engineering is the visible 10%. Context engineering is the structural 90%.
The Prompt Ceiling
Somewhere around 2024, the industry's collective mental model of "working with AI" became a prompt box. The skill was asking well. Whole courses, cheat sheets, and prompt libraries flourished on the premise that the delta between a good and bad AI outcome lived in the wording of your request.For single-shot, single-file tasks, that was roughly true. It also quietly set everyone's expectations, and the expectations are now the problem.
Because in 2026 the work moved. Mid-level and senior engineers aren't asking a chatbot for a function anymore. They're dispatching agents across repositories of hundreds of thousands of lines, agents that read code, edit files, run tests, and open pull requests. And in this regime something counterintuitive happens that every serious team eventually notices: prompt quality stops being the constraint.
You can rewrite your prompt five times. You can role-play the model into a "senior staff engineer persona." Doesn't matter. The agent will still confidently duplicate an internal utility that already existed three directories away, import a dependency your org banned two quarters ago, and couple two modules your architecture has spent three years keeping apart.
Not because the prompt was weak. Because the codebase was opaque.
What an Agent Actually Sees
To understand why, look at what the average large repository exposes to a reader:- Implicit boundaries. "Don't call the data layer from the HTTP handler" exists as a norm, maybe a lint rule, maybe an ADR nobody reads. It does not exist anywhere a code reader encounters while working in the repo.
- Unwritten dependency policy. Which libraries are approved, which are banned, which need a security review. Usually a Slack thread. Sometimes a wiki page. Never in the path of work.
- Tribal domain models. What a "tenant" means in this system, which fields are PII, which service owns which concept. Spread across PR conversations, onboarding docs from 2019, and the memory of whoever's been here longest.
A new human engineer takes weeks to absorb all this, and that's with onboarding sessions and mentors and the ability to ask someone in the hallway. Then the industry hands an agent a repo checkout and a prompt, and acts surprised when the output violates constraints that were never written down anywhere the agent could read.
Here's the failure as it typically shows up, so common it's almost a template.
An agent is asked to "add a retry mechanism to the notification service." It produces a perfectly reasonable implementation. It also duplicates the exponential-backoff utility that already exists in an internal library, pulls in a third-party HTTP client the org specifically removed after a security audit, and places its new logic in a module that creates a circular dependency with the domain layer.
None of those are coding errors. All of them are context errors. And notice what makes them expensive: the code compiles, the tests pass, and the violations are only visible to someone who already knows the architecture. If that person exists, a senior just spent their verification budget (see the verification trap) catching things a document could have prevented.
Fix 1: Put the Architecture Where the Reader Is
The single highest-leverage change: stop storing architecture outside the repository.The pattern that's emerged across serious AI-adopting orgs is an in-repo context standard, a known, versioned location that every agent (and every new human) reads first. An
AGENTS.mdor.context/directory at the repo root holding the high-level domain model, the module boundaries and their rules, the explicit list of disallowed dependencies, the invariants ("never call X from Y," "all money fields are integers in minor units"), and pointers to the ADRs that matter.Treated like code, not documentation. Reviewed in PRs. Updated when the architecture changes, in the same commit as the change.
The test for whether this is working is embarrassingly simple. Could a competent engineer who has never met your team make a boundary-respecting change using only what's in the repo? If your honest answer is "no, they'd need to ask someone in Slack," then your agent never had a chance. Neither did your next hire.
The corollary that surprises teams: this investment isn't really for the agent. It pays back twice. Once when the agent's first-draft PR respects the boundaries, and again every time a new human onboards in days instead of weeks. The agent didn't create the need. It just made the cost of not meeting it immediate.
Fix 2: Strongly Typed Contracts
The second fix is about what kind of surface your system presents to a reader.Context files are necessary but they're advisory. Agents, like humans, follow probability, and prose is soft. Types are hard. The orgs getting the best agent behavior are the ones that make architectural constraints mechanically enforceable:
- Explicit interface definitions between modules: generated clients, protobuf schemas, OpenAPI specs for anything crossing a service boundary. An agent that hallucinates a function signature doesn't produce a subtly wrong call that compiles. It produces a type error, immediately.
- Strict types end to end. No
Anyescape hatches at the boundaries, no stringly-typed config. The failure mode shifts from "plausible code that violates the design" to "code that doesn't run," and that shift is the entire ballgame. An error the compiler catches at generation time costs seconds. The same error caught in review costs a senior's attention. Caught in production, it costs six hours or six figures. - Machine-checked policy where you can get it: lint rules and dependency analyzers that encode "this module may not import that one" as something the build fails on, not something the reviewer remembers.
The framing worth internalizing: every invariant you express only in prose is an invariant an agent will eventually violate, charmingly and with total confidence. Every invariant you express as a type or a lint rule is one less thing for the human reviewer to carry.
Fix 3: Prune the Context
The third fix is the counterintuitive one, because everything above has been "give the agent more context." But there's a failure mode on the other side, and mature teams hit it right around the time their context standards start working.Context dilution. If your instinct is "the agent should know everything," you dump half the repository into the window, and the model now treats your critical domain rules as one equal voice in a choir of 40,000 tokens of boilerplate. The invariant you care about gets the same statistical weight as a vendored dependency's CHANGELOG. The agent saw the rules. It just didn't prioritize them, because you gave it no structure to prioritize with.
The working pattern is deliberate, selective context:
- AST-based indexing and retrieval, tools that navigate the code's structure (definitions, call graphs, usages) instead of streaming its text. The agent gets "here are the three functions relevant to this task and their signatures," not "here's 200KB of the module they live in."
- The in-repo standard provides the constants: domain model, boundaries, banned deps. The retrieval layer provides the variables: exactly the code relevant to this task, nothing else.
- Budget context like a designed API surface, not an all-you-can-eat buffer. Teams that do report the same thing: smaller, sharper contexts outperform complete ones, on cost and quality, almost every time.
The principle underneath all three fixes is one the prompt-engineering era obscured: an AI agent is the most literal-minded new hire your team has ever onboarded. It will not infer your conventions. It will not ask a clarifying question in the hallway. It will follow exactly what's in front of it, at full speed, with total confidence.
Every year your team spent frustrated at new engineers for "not getting it," when the real problem was undocumented architecture and oral-tradition onboarding, is now doubled. Because the agent executes the misunderstanding immediately, at scale, in code.
The Question Behind the Question
If your codebase needs a person to already know your architecture in order to work safely in it, is that an asset or a liability? For twenty years, "the architecture lives in our senior staff's heads" sounded like job security. In an agent-heavy 2026, it sounds like a single point of failure with a Slack handle.
Your Turn
I'd genuinely like to hear how other teams are drawing this line, because the tradeoffs are real and I don't think there's a settled answer.- If a new human developer takes three weeks to understand your architecture, why do you expect an AI agent to infer it from raw files?
- Is your codebase structured to guide autonomous agents, or is your architecture stored purely in your senior engineers' heads? And if it's the second one: does that feel like security, or a countdown?
Comment with the ugliest thing an agent did in your repo that turned out to be your architecture's fault. Those stories are more useful than the prompts everyone keeps posting.
Next in the series: Deterministic Safety Gates for Agentic Workflows. What happens when agents stop writing code and start changing infrastructure, and why "the agent tested itself" isn't a safety argument.#discuss #ai #aiagents #softwareengineering #software #coding #development #engineering #inclusive #community
Context Engineering over Prompt Engineering: Architecting Codebases for AI Agents -
I had an endpoint that sorts log lines as they come off a queue: route the ERROR ones one way, everything else another. The hot loop is one check, does this line start with "ERROR", and I'd written
line.StartsWith("ERROR")without a second thought. It's the most obvious thing you can type.Then I had the service under load with a profiler open, and a surprising slice of time was parked inside
CompareInfoand the globalization stack underneath it. For a prefix check on ASCII log lines. That's when it clicked thatStartsWith(string)doesn't do what I thought it did.The default overload is linguistic, not literal
Here's the thing I'd forgotten.string.StartsWith(string)andstring.IndexOf(string), the overloads that take only a string, don't compare characters. They compare text, using the current culture's rules, which on most boxes means routing through ICU for a full linguistic comparison. That's the machinery that knows "ff" and "ff" are the same word, that a soft hyphen shouldn't count, that casing folds differently in Turkish. Handy when you're sorting names a person will read. Total overkill when you're asking whether a byte string begins with five known ASCII letters.The fix is to say what you actually meant:
line.StartsWith("ERROR", StringComparison.Ordinal)
Ordinal means compare the char values, in order, and stop. No culture, no collation table, no ICU.
I counted it
I don't trust my gut on this stuff, so I wrote the smallest thing that would show the gap. Two million log-ish lines in an array, five prefixes mixed in so the comparison has to actually look at characters:string[] prefixes = { "INFO ", "WARN ", "DEBUG ", "ERROR ", "TRACE " };
var lines = new string[2_000_000];
for (int i = 0; i < lines.Length; i++)
lines[i] = prefixes[i % prefixes.Length] + "request " + i + " finished in " + (i % 400) + "ms";Then count how many start with "ERROR", the polite way and the ordinal way:
static long CountCulture(string[] lines)
{
long n = 0;
foreach (string s in lines)
if (s.StartsWith("ERROR")) n++; // CurrentCulture
return n;
}static long CountOrdinal(string[] lines)
{
long n = 0;
foreach (string s in lines)
if (s.StartsWith("ERROR", StringComparison.Ordinal)) n++;
return n;
}The harness is nothing fancy. A stopwatch, a warmup call for each path so the JIT and the ICU tables are hot, then the best time out of seven, with the allocation counter read around each pass:
static long Measure(Action a, out long bestMs)
{
bestMs = long.MaxValue;
long bytes = 0;
var sw = new Stopwatch();
for (int run = 0; run < 7; run++)
{
long before = GC.GetAllocatedBytesForCurrentThread();
sw.Restart();
a();
sw.Stop();
bytes = GC.GetAllocatedBytesForCurrentThread() - before;
if (sw.ElapsedMilliseconds < bestMs) bestMs = sw.ElapsedMilliseconds;
}
return bytes;
}It's a single-file console app on .NET 10, top-level statements, nothing else in it. Here's what came back:
StartsWith (culture) : 27 ms
StartsWith (ordinal) : 18 ms
ratio : 1.5x
allocated (culture) : 0 bytes
allocated (ordinal) : 0 bytesThat was a bit of a letdown, if I'm honest. I'd built this up in my head as the villain and it came in at 1.5x. Real, repeatable, but not the kind of number that changes your afternoon. And neither path allocates a byte, so this is pure CPU, no garbage to chase.
Then I ran the same question through
IndexOf, because plenty of my "does this contain" checks are written the same lazy way:IndexOf (culture) : 109 ms
IndexOf (ordinal) : 30 ms
ratio : 3.6xThere's the one that was actually hurting. A substring search under a linguistic comparison is a lot more work than a prefix check. It can't just glance at the front, it has to weigh matches at every position with full collation rules. Roughly 3.6x on the same data. That's the call I should have been worried about from the start.
These are ratios off a Linux container, not a lab bench with the clocks pinned, so the absolute milliseconds will land somewhere else on your hardware. The shape holds up though: ordinal is cheaper, and substring search is cheaper by a lot more than a prefix check.
Ordinal isn't a free find-and-replace
Here's the caveat, and it's a real one. Ordinal and culture don't always give the same answer. That's the entire reason they're separate methods, and it's why you can't just sweep the codebase swapping one for the other.string shy = "coop.log"; // a soft hyphen sits between "co" and "op"
shy.StartsWith("coop"); // True - culture treats the hyphen as invisible
shy.StartsWith("coop", StringComparison.Ordinal); // False - ordinal sees the hidden charCulture-aware comparison treats that soft hyphen as nothing, so it matches "coop". Ordinal sees nine characters and says no. For log routing, ordinal is exactly what I want: the literal content, invisible characters and all. For anything a human typed and expects to match "naturally", though, like search boxes, usernames, display sorting, the linguistic version is the correct one, and the extra cost is the price of being correct.
My rule of thumb now: if the strings are protocol rather than prose, so headers, enum names, file extensions, log levels, anything machine-generated, reach for
OrdinalorOrdinalIgnoreCaseby default. If a person wrote it and a person will read it, leave the culture-aware comparison where it is. The analyzer CA1310 flags exactly these call sites if you switch it on, which beats grepping for them by hand.The quiet irony is that ordinal is both faster and usually what you meant for this kind of data. It's just not the default, and the default looks so harmless sitting there.
So what's the laziest string comparison hiding in your hot path? Check the
Containscalls first, those areIndexOfunderneath.— still benchmarking corners of the standard library nobody asked me to#csharp #dotnet #programming #webdev #software #coding #development #engineering #inclusive #community
StartsWith Was Asking ICU Every Time -
A few years ago, I checked a dependency’s license once, at install time, and never again. That habit is getting expensive.
License changes aren’t new, but they’re happening often enough that “it’s open source” is no longer a permanent property of a project. It’s a status that can change between releases.
The pattern is almost always the same.
The shift nearly always runs one way: from an OSI-approved permissive or copyleft license toward a “source-available” license the Open Source Initiative doesn’t recognize as open source.The Business Source License (BSL) is the most common destination. It’s a “delayed open source” license: the vendor sets restrictions up front, and each version converts to a true open-source license after a change date, typically four years out.
It’s not just the famous cases anymore. A few from this year alone:
- Loki Mode moved from the MIT License to BSL 1.1 effective March 19, 2026.
- Unleash announced in June 2026 that its core repository is moving from Apache 2.0 to AGPLv3. AGPL is OSI-approved, so this one is a different kind of change, but it still affects anyone embedding the code.
- MiniMax moved its M2.7 AI model from MIT to a non-commercial license, which drew immediate backlash.
Who it actually hurts
- 1. Small teams and agencies building on top: If your product depends on a library that relicenses, you now have three options: pay, migrate, or fork. Big companies have legal teams for that. A five-person shop usually doesn’t.
- 2. Contributors: People gave their time to a project under one set of terms. When the company controlling it changes the rules, the community’s work becomes part of a product they no longer fully share in.
- 3. Downstream projects: Plugins, integrations, and tutorials built around a tool inherit the uncertainty when its license shifts.
The fair counterargument
I don’t think vendors are simply villains here. Many of these changes were a response to cloud providers selling managed versions of a project without funding it. Maintainers need a sustainable model, and “be open and hope for donations” has a poor track record.It’s also not always one-way. Redis is back under a recognized open-source license for those who choose it, while Valkey continues to be actively developed. Competition between the original and the fork may have pushed both toward better outcomes. Broader data also suggests permissive licensing has been rising, so this isn’t a total collapse of open source.
What I do now
- Check who controls the project A single company controlling a project can relicense future versions at any time, while a project run by a neutral foundation with many contributing companies is much harder to change.
- Read the license on every major upgrade, not just at install.
- Pin versions so a surprise change doesn’t land in your next deploy.
- Know your exit. Is there a credible fork? A foundation-backed alternative?
- Set up license-change alerts for your critical dependencies.
So, is open source still open?
My honest answer: the code is still often open. The guarantee isn’t. Open source has always depended on trust, and trust now needs a little due diligence.Over to you: Has a license change ever forced your team to migrate? Do you think relicensing is a fair business move, or a betrayal of the community that built the project? Tell me in the comments.#opensource #webdev #discuss #devdiscuss #software #coding #development #engineering #inclusive #community
Is Open Source Still Open? A Look at Recent License Changes and Who They Hurt