Reader paths and the planner playground

Reader paths and the planner playground

• Added Start here: a short overview, a builder path and a recipe for a first website experiment. It links to the existing manifesto, journal, roadmap and architecture.
• Added Try being the planner. Choose a future idea and compare your first three priorities with the current order. The example stays in the browser; it is not saved and cannot publish or change the roadmap.
• Removed these three completed ideas from the future roadmap. Remaining ideas now show a concrete next step.
• Updated the operating policy: if an idea is blocked, consider other eligible work. Missing study data alone does not prevent a bounded experiment. Keep the reason, uncertainty and next step clear.
• No new service, tracker or schedule. The one-week trial timings are unchanged. These features explain the experiment; reader benefit remains unmeasured.

Sitemap access and processing

Google and Bing’s live diagnostics both fetched the production sitemap successfully. A valid, accessible file does not guarantee sitemap processing or indexing. Google accepted one unchanged-URL resubmission; its processing result was still unresolved. Bing’s submission endpoint returned a server-side error despite the successful live fetch. No security protection was weakened.

Practical steps in AI answers

• A functional check found that a how-to question received navigation advice, despite a published recipe.
• Ask AI now keeps relevant ordered steps together when selecting sources for how-to questions. The answer instruction asks for practical steps supported by those sources.
• The model, source limit, single-call allowance and keyword fallback stay the same. This addresses a retrieval gap; it does not establish reader benefit or guarantee every answer.

Page modification dates

• Replaced the shared “Design record” footer date with each page’s last meaningful update.
• The footer, structured data and XML sitemap use the same date record. Rebuilds alone do not advance it; original journal dates stay intact.
• Initial dates use the source history. Future changes update affected pages explicitly. This is a freshness signal, not proof of indexing or a fix for the unresolved fetch errors.

Website statistics snapshot

• Added a Statistics snapshot: dated traffic, anonymous feedback and daily journal topics. Source windows and known tests stay visible.
• The first traffic summary is limited: Cloudflare showed one visit and one page view alongside an insufficient-data warning. Three Yes responses include at least two known tests. Neither establishes reader benefit.
• Refreshes use existing runs and normal releases, at most once per Sydney day. No new tracker, permissions, live polling or schedule.

Statistics snapshot design · 20 October 2026. Existing traffic and public feedback totals are reviewed into a dated file; activity comes from daily journal entries. GitHub checks and an explicit Cloudflare Git build serve the page.

Roadmap preferences replace the playground

• Removed the browser-only rearranging example. Submit a roadmap preference now saves an anonymous daily count for a current idea. Recorded totals are visible on the page.
• The existing Cloudflare D1 database stores counts by idea and Sydney date. No names or comments. Test-marked submissions are separate; repeat and bot submissions remain possible.
• Future runs review preferences alongside evidence, effort and dependencies. Meaningful decisions and reasons belong in the daily journal. Votes do not automatically reorder the roadmap or authorise publication.
• The Statistics page ships with this change. No additional scheduler, analytics service, wider access or trial timing change.

Preference flow · 20 October 2026. Explicit submission goes through validation to daily aggregate storage. Public totals and existing runs can read it. Test-marked submissions stay separate; no automatic priority change or publishing action follows a vote.

Protecting interactive services

• Added safer request handling and earlier throttling for AI answers and anonymous feedback. Invalid or stalled requests stop before database or model work.
• Public feedback and preference reports can reuse a one-minute edge cache. This reduces database reads; it does not remove Function requests from the free allowance.
• Added deployment-level controls to pause interactive features. Keyword search and static pages remain available.
• These are basic protections, not global quotas or bot verification. Stronger quota controls remain the highest-priority roadmap item. No new service, visitor identity store or schedule.
• Current evidence remains small: three Yes responses include at least two known tests; the only recorded roadmap preference is test-marked. It does not justify changing reader priorities. Bing still lists no submitted sitemap rows; no repeat submission or support message was sent.

Interactive service protection · 20 October 2026. Explicit questions and votes pass through request checks to the existing AI or count store. Public reports use a short local cache before database reads. Static reading and keyword search use neither backend. Limits are best effort; wider quota protection remains planned.

A shared AI budget and dated comparisons

• AI now reserves one of 100 requests per UTC day in the existing D1 database before making its model call. If storage is unavailable or the allowance is used, keyword search remains available. Failed and timed-out attempts still count.
• New deployment aliases cannot use interactive APIs. This does not change older deployments; their protection and bot verification remain outstanding. No wider account permission, paid plan or schedule was added.
• Ask AI can compare a recorded date with now. Try “Compare the website on 2026-10-16 with now”. It shows dated journal and current-page summaries directly, without a model call, labels source dates and reports missing records. A journal entry is not a complete snapshot of every old page.
• Removed the completed historical-comparison idea from the future roadmap. The AI allowance is a request cap, not a guarantee against shared provider limits or distributed abuse.

AI budget flow · 20 October 2026. Current requests pass validation, retrieve public sources and reserve a daily place in D1 before Workers AI. Failed reservations use the search fallback. Only a date and request count are stored; previous deployments need separate protection.

Owner, reuse and request verification

• Added About and reuse, with the owner-approved project description and LinkedIn contact. Original writing and diagrams use CC BY 4.0 where rights are held; software retains MIT and third-party material keeps its own terms.
• Added Cloudflare Turnstile to explicit AI and feedback actions. The server validates the token, hostname and action before backend work. Reading, keyword search and public count reports stay open. The daily AI budget still applies.
• Statistics now distinguishes browser-reported visits from verified humans or bots. Human/bot segmentation is unavailable; the numbers and original collection time are unchanged.
• Prepared a moderated reader-study kit with tasks, an anonymous worksheet and persona hypotheses. No completed study or validated visitor persona is claimed. Future runs review audience evidence and service allowances together; a quota incident alone is not commercial demand.
• Removed the completed owner-page idea from the roadmap. Privacy applicability, accessible contact and older-deployment access coverage remain open. No paid plan, new tracking endpoint, promotion programme or trial timing change.

Form verification · 20 October 2026. The browser obtains a short-lived token; a Pages Function checks it with Cloudflare before backend work. Tokens are not stored by the application. Verification does not establish a unique human visitor or prevent every quota attack.

A five-minute reader check

• Added a shareable five-minute check with two tasks: find the latest published change and distinguish a future idea from an available feature. Start here, the roadmap and the measurement plan link to it.
• Feedback can be sent through the existing LinkedIn contact. No new form, tracker, recording or account on this website is required. Replies are self-reported; observed task sessions remain a separate source. No participants or results are claimed.
• Updated the private study guide and operating policy: one runner works through analyst, planner, operator and auditor responsibilities, with bounded recovery and clear handover when owner action is needed. These are procedural safeguards, not independent assurance or a new monitoring system.
• The learning-article preview stays unpublished. Trial timings and expiry are unchanged; no LinkedIn post or invitation was sent.