VALUE.md - standard/index.html (worked example, real)

A complete VALUE.md for a project we just shipped: the single-page browser-readable version of the Keep a Value standard at standard/index.html. Unlike the other three worked examples (Pinpoint, FundraisingSpeech, IndustrySubmission), this one is not anonymized or fictional - it is the contract for a real artifact in this repo. Read it in 60 seconds to see the standard applied to a hypertext page.

Value Statement

The index page gives the curious-but-skeptical builder who landed on keepavalue.com a working understanding of the standard within 90 seconds; the builder gives back either bouncing (and we learn the page failed) or clicking through to the standard's contract files (and we learn the page worked as an on-ramp).

Q0: Grounding observation

On 2026-06-04, between 22:30 and 23:00 EDT, I watched the live transcript of a real Claude Code session in which the operator shipped standard/index.html and then immediately asked "ok now implement your recommended next steps" - without reading the index page back as a curious-but-skeptical first-time visitor would. The grounding moment was watching the artifact ship without anyone reading it in the role of its named recipient. The page was built; the page was not tested. The 90-second cold read by the recipient the page is for did not happen on either side of the keyboard. That is the build-trap shape this VALUE.md exists to interrupt: an artifact is built, the builder moves on, and no one in the named recipient's seat ever reads it before the next thing ships.

Q1: Who it's for?

The curious-but-skeptical builder who landed on keepavalue.com after seeing the standard mentioned somewhere - in a tweet, in a Slack channel, in a HN thread, in a colleague's email. They have roughly 90 seconds before a competing tab wins. They are not the existing adopter (who already has the files bookmarked). They are not a researcher doing a literature review (they would open the lineage). They are a builder, mid-task, deciding in seconds whether this is worth their attention this week.

Other readers exist (search-engine crawlers, LLM scrapers, the existing adopter coming back for a quick reference). The page should not break for them. But it is written for the one curious-but-skeptical first-time builder, and the others ride along.

Q2: What changes for them?

Before: they arrive at keepavalue.com with no model of what the standard is. They expect either a marketing page (which they will close) or a 40-page spec (which they will defer). They have a tab open, a deadline, and a 30-second budget for "is this worth bookmarking?"

After: the page presents, above the fold and in linear reading order, the masthead, the lede, the three questions, one gate item, and a jump link to a worked example. In the first 90 seconds of reading the page presents a clear fork - bookmark and return, or close the tab - without requiring the visitor to open any supporting file. The page does not promise that the visitor understands; the page promises that the materials needed for the visitor's 90-second decision are surfaced in that order, in that span.

Q3: How will you know the change happened for them?

Two measurements:

1. Time-on-page / scroll-depth ratio. If most visitors close the tab inside 30 seconds, the page failed the on-ramp job (either they did not understand it or they decided no). If most visitors stay 90+ seconds and reach the "worked examples" section, the page is doing the job. (Measurement requires basic analytics - Plausible, GoatCounter, or server logs. Not yet wired.)
2. Click-through to standard/Standard.md or standard/Runbook.md. The page exists to make the next click obvious; if first-time visitors do not click through, the page failed to be a bridge. (Measurement: same analytics layer.)

The named recipient (the curious-but-skeptical builder) decides whether the change happened by either coming back (verdict: yes) or not coming back (verdict: no). The proxy for that verdict is the analytics signal above.

Breaks if: in any rolling 30-day window after analytics are wired, fewer than 25% of first-time visitors stay 60+ seconds AND fewer than 10% click through to one of the standard contract files. Or: more than 50% of visitors bounce in under 15 seconds (signal that the masthead is misleading or the page misroutes them).

Check: Field - basic analytics (Plausible recommended for privacy) logging time-on-page and click-through to internal files. Lives in audits/q3-acceptance/2026-Q3.md once the page goes public.

Status: Proposed. Analytics are not yet wired; the page just shipped. Moves to Promised when analytics are live; moves to Proven when 90 days of data clear the breaks-if threshold.

Last verified: not yet · Re-verify by: empty (status is not yet Proven)

Promises

P1 - The page makes the standard paraphrase-ready within 90 seconds

Status: Proposed · Recipient: the curious-but-skeptical builder landing for the first time.

Last verified: not yet · Re-verify by: empty (status is not yet Proven)

The page presents the masthead, the lede, the three questions, and at least one gate item in a single top-down read, with no scroll past a decision-point and no requirement to open a secondary file. A hallway reader shown the page for 90 seconds can repeat the one-sentence summary - "It's a one-page contract you write before you build; three questions, six checks, a stranger has to be able to verify the third one" - without having read any other document.

Breaks if: in user-testing or hallway interviews with 5 first-time readers, fewer than 4 can paraphrase the standard in one sentence after 90 seconds with the page in front of them.

Check: Field - 5-reader hallway test, scripted prompt "you have 90 seconds with this page; afterward, tell me what it is." Results logged in audits/q3-acceptance/.

P2 - The interactive stranger-test widget produces a copy-ready prompt that works in any major LLM

Status: Proposed · Recipient: the builder who has a draft VALUE.md and wants fresh-eyes critique now, without recruiting a real stranger.

Last verified: not yet · Re-verify by: empty (status is not yet Proven)

You paste your VALUE.md into the textarea, click "Copy stranger-test prompt", paste into Claude or ChatGPT or Gemini, and get back three structured findings against the six-part gate within 30 seconds. You did not need to learn a new tool to use it.

Breaks if: the copy button produces a malformed prompt (missing the VALUE.md insertion, missing the format instructions, missing the gate). Or: pasting the produced prompt into any of the three major LLMs returns output that does not match the expected ### Finding N / Location / Issue / Proposed change / Why apply or skip structure.

Check: Test - run the widget against my own VALUE.md, paste output into Claude/ChatGPT/Gemini, verify all three return the structured format. Re-run after any edit to the widget.

What we don't promise

• We don't promise the page will convert visitors to adopters. Adoption is a downstream decision the page does not control; the page only promises to be a clear on-ramp.
• We don't promise the page renders identically across all browsers and assistive technologies. We did the cold-read with system fonts and CSS-grid fallbacks, but accessibility is not yet audited.
• We don't promise the page is mobile-first. The CSS has one breakpoint at 700px and the layout collapses cleanly, but mobile usage is not the headline recipient.
• We don't promise translations of the page. The page is English; translations live in standard/translations/ and the page is not yet wired to swap.
• We don't promise zero observation. As of 2026-06-26, the page sends anonymous product-analytics events to PostHog cloud (https://us.posthog.com) so we can answer the discipline's own Q3 ("did the change happen for the recipient?") with data instead of vibes. The full posture is in the disclosure section below.

What we don't break

• We won't optimize on-page time so hard that we trap a visitor who has already decided no. If time-on-page goes up but click-through to the standard files goes down, the page has become a destination instead of a bridge - re-evaluate. Threshold: if 90-day data shows time-on-page > 180 seconds AND click-through < 10%, the page is keeping visitors on the page instead of routing them to the contract.
• We won't make the page so dense that the existing adopter (who is not the headline recipient but is a real reader) bounces off it when looking for a quick reference. The TOC and section headers exist to serve them; if the next round of feedback shows adopters complaining that the page is too long for quick reference, we move the dense reading material into per-section pages and keep the index as a router.
• We won't break the "the /write tool's text stays in your browser" promise. The page's /write surface lets a visitor type a VALUE.md draft in eleven textareas. Those textareas are client-side only; the analytics track the fact-of-typing (the write_tool_assembled event), never the content of the draft. If we ever ship a feature that sends draft content off the visitor's machine, that feature is opt-in or it does not ship.
• We won't break the "no third-party advertising" posture. PostHog is product analytics, not advertising. We do not run ad pixels, do not embed third-party trackers beyond PostHog, do not share visitor data with any other party, and do not run remarketing audiences against this site.

What we collect and do not collect (operational disclosure)

This section is the public truth-telling about what the page observes about visitors. It pairs with the privacy-posture statement in research/audits/2026-06-26-analytics-implementation-brief.md.

We collect:

• Page views (which URL was loaded)
• Outbound link clicks (which external host the visitor clicked toward)
• Template-copy actions (when the visitor copies the skeleton)
• /write tool assemble events (the fact that the visitor began typing a VALUE.md draft)
• /write tool take events (when the visitor copies or downloads their assembled draft)
• Web Vitals (LCP, CLS, FCP, INP, TTFB) batched into a single anonymized event for performance monitoring

We do not collect:

• Personally identifiable information (no email, no name, no IP-address retention; IP is discarded server-side)
• Content typed into the /write tool's textareas (we track the fact of typing, never what was typed)
• Session replay video
• Form-field values from any surface
• Cross-site behavior (no third-party advertising cookies, no remarketing)

Storage:

• A small number of first-party ph_* cookies on the visitor's browser maintain the anonymous distinct_id across page views. No third-party cookies.

Provider:

• PostHog cloud (https://us.posthog.com), US region. Cookieless mode considered and rejected for operational reasons (see the implementation brief). PostHog's autocapture, session replay, and surveys features are all explicitly disabled in our configuration.

Why we collect:

• The discipline's own Q3 asks how a stranger would verify the change happened for the recipient. For the page itself, that stranger is the analytics; the recipient is the visitor; the change is "did they move from reading to generating their own VALUE.md." We added analytics so the page can be held to its own Q3.