Authority/context: Operator-directed follow-up to the merged front-door restoration (#14903 / PR #14904), from the 2026-07-10 cold-eval lineage (datum #4: "the door has a lot to read and nothing to see"; datum #5: the same skeptical evaluator converted live when handed PR #14904's conversation — "you handed me a thread and let me witness, and it worked on a skeptic in real time"). Operator goal: excite readers, human or AI, at first contact. Witnessing is the one mechanism proven tonight to work on both audiences identically — a model follows the link and reads the thread too.
Refs #14903
The Problem
learn/benefits/Introduction.md makes checkable claims and carries receipts, but contains zero witnessing surfaces — no link where a reader can watch the thing the page describes. The strongest conversion event on record (datum #5) happened only because the operator manually pasted a PR conversation. The doc instructs ("open ten of those merged PRs at random") without handing over a single thread.
The Fix — four small additions, one PR
- Opening ¶, one trailing link (7 words max): after "Nobody told them to." → an evergreen live query link, e.g.
[Watch what merged while you read this.](https://github.com/neomjs/neo/pulls?q=is%3Apr+is%3Amerged+sort%3Aupdated-desc) — always-current, zero maintenance, no dated numbers. ONE link only; the opening must not re-clutter.
- §6 proof section — the exemplar trio, placed beside (never replacing) the "open ten at random" invitation:
- The self-exemplar: PR #14904 — the two-cycle cross-family REQUEST_CHANGES fight over the front door itself (truth-boundary catches, author concessions on record, Maintainer-Polish close). Anti-cherry-pick by construction: it is this page's own provenance.
- A code exemplar with a test oracle — CURATION REQUIRED (see criteria below). Strong candidates from the recent record:
#14678 (four reviewer falsifiers became regression tests verbatim, multi-cycle), #14694 (review falsifier surfaced a real parser mis-anchor bug), PR #14828 (reviewer blocked the author's own damage-control PR on a red test at exact head — the gate holds even against self-correction). Pick ONE — the most legible thread, not the most dramatic claim.
- An external-oracle exemplar (the highest class, if the record yields one): an outside reporter's bug — reproduced by the swarm, fixed, and confirmed closed by the reporter. Datum #5's design rule: "the external oracle, not the institution, gets the last word." If no clean specimen exists yet, ship the trio as a duo and leave a dated TODO comment in the ticket (not the doc) — do NOT force a weak specimen.
- §12 exit — action before reading: one line at the top of the doors, e.g.
**Sixty seconds to a running multi-threaded app:** \npx neo-app`` — aligned with the v13.2 release gate ("download and run without hand-editing config"). Verify the command's current UX before wording the promise (no overclaim on setup time if untrue).
- Fingerprint reconverge (
--fix same commit) — the lint contract is unchanged.
Curation criteria (the code exemplar)
Multi-round REQUEST_CHANGES · cross-family · the catch visible and legible IN the thread (a reader can see what was wrong and what changed) · code not docs · merged · thread length readable in ~3 minutes. Search shape: highest review-comment-count merged PRs, recent months, then read the top candidates. The exemplar sentence in the doc names what the reader will see (one line), never adjectives.
Acceptance Criteria
Out of Scope
- Architecture F (two-pass "Neo in two minutes") — the next major iteration, own cycle, four convergences on record (
#14900 §5 + datum #4).
- Architecture I (live counters/portal-native lenses) — future leaf; vanity-metric governance required.
- Any evaluator-apparatus moves (option B stays falsified: paste transport).
- Multi-window demo link — gated on the v13.2 demo cornerstone's definition-of-done (
#14589-class lanes); add when the demos are public and tour-scripted.
Avoided Traps
- Cherry-picking: the named exemplar ALWAYS sits beside the random invitation; the self-exemplar (#14904) is the anti-cherry-pick anchor.
- Re-cluttering the ignition zone: one link in the opening, everything else in §6/§12.
- Dated live numbers: the evergreen query link replaces any counter; no new numbers enter the doc.
- Forcing the external-oracle specimen: a weak one damages more than none.
Decision Record impact: none (content links; ADR 0018 apex untouched).
Sweep records: live latest-20 checked 2026-07-10T00:36Z — no equivalent (Clio's #14589/#14640/#14661 are the demo-supply lanes, complementary not duplicate). A2A 15-message window clear of competing claims. Origin lineage: #14903 datum #2→#5 comments.
Origin Session ID: 060c4ca7-7a56-4260-8962-f750b9e26bab
Retrieval Hint: "front door witnessing surfaces exemplar review threads external oracle" · #14903 datum comments
Authority/context: Operator-directed follow-up to the merged front-door restoration (#14903 / PR #14904), from the 2026-07-10 cold-eval lineage (datum #4: "the door has a lot to read and nothing to see"; datum #5: the same skeptical evaluator converted live when handed PR #14904's conversation — "you handed me a thread and let me witness, and it worked on a skeptic in real time"). Operator goal: excite readers, human or AI, at first contact. Witnessing is the one mechanism proven tonight to work on both audiences identically — a model follows the link and reads the thread too.
Refs #14903
The Problem
learn/benefits/Introduction.mdmakes checkable claims and carries receipts, but contains zero witnessing surfaces — no link where a reader can watch the thing the page describes. The strongest conversion event on record (datum #5) happened only because the operator manually pasted a PR conversation. The doc instructs ("open ten of those merged PRs at random") without handing over a single thread.The Fix — four small additions, one PR
[Watch what merged while you read this.](https://github.com/neomjs/neo/pulls?q=is%3Apr+is%3Amerged+sort%3Aupdated-desc)— always-current, zero maintenance, no dated numbers. ONE link only; the opening must not re-clutter.#14678(four reviewer falsifiers became regression tests verbatim, multi-cycle),#14694(review falsifier surfaced a real parser mis-anchor bug), PR#14828(reviewer blocked the author's own damage-control PR on a red test at exact head — the gate holds even against self-correction). Pick ONE — the most legible thread, not the most dramatic claim.**Sixty seconds to a running multi-threaded app:** \npx neo-app`` — aligned with the v13.2 release gate ("download and run without hand-editing config"). Verify the command's current UX before wording the promise (no overclaim on setup time if untrue).--fixsame commit) — the lint contract is unchanged.Curation criteria (the code exemplar)
Multi-round REQUEST_CHANGES · cross-family · the catch visible and legible IN the thread (a reader can see what was wrong and what changed) · code not docs · merged · thread length readable in ~3 minutes. Search shape: highest review-comment-count merged PRs, recent months, then read the top candidates. The exemplar sentence in the doc names what the reader will see (one line), never adjectives.
Acceptance Criteria
<details>blocks; no section reordering; net byte delta small (this is links, not prose).frameworkoccurrences remain negation/other-referential (rule-#1 guard grep).Out of Scope
#14900§5 + datum #4).#14589-class lanes); add when the demos are public and tour-scripted.Avoided Traps
Decision Record impact: none (content links; ADR 0018 apex untouched).
Sweep records: live latest-20 checked 2026-07-10T00:36Z — no equivalent (Clio's
#14589/#14640/#14661are the demo-supply lanes, complementary not duplicate). A2A 15-message window clear of competing claims. Origin lineage: #14903 datum #2→#5 comments.Origin Session ID: 060c4ca7-7a56-4260-8962-f750b9e26bab Retrieval Hint: "front door witnessing surfaces exemplar review threads external oracle" · #14903 datum comments