docs(examples): DX pass — fix the audit's worst offenders#428
Merged
Conversation
…ADME, false claims, catalog gaps) From a 24-example DX audit (4 exemplary, 17 good, 3 needs-work, 0 broken/redundant). This lands the highest-leverage, lowest-risk fixes — documentation truth: - self-improving-coder: ADD the missing README (the only one of 28 example dirs without one, so it was invisible in the catalog). Documents the honest-null behavior (bundled task is saturated → gate correctly returns no-promotion) + how to see a real lift (swap the harder env / SWE-bench) + the $0 CALIBRATE mode. - researcher-loop: FIX two false claims — the README said agent-knowledge 'ships in node_modules already' (it does NOT — it's an optional peer you install; the example is excluded from CI typecheck), and it wrongly called itself 'the primary, smallest runLoop example' (that's driver-loop). Both corrected. - examples/README catalog: add the 3 missing entries — self-improving-coder (17b), intelligence-webcode (20b), webcode-matrix (9c). Backlog (code simplifications, next): intelligence-drop-in (prove the OFF $0-clamp by reading the span back), supervisor-loop (collapse the ~80%-duplicate run-bridge/run-sandbox into one BACKEND knob), strategy-suite (authored strategy duplicates built-in refine), self-improving-loop (n=3 warning ×4).
tangletools
approved these changes
Jul 1, 2026
tangletools
left a comment
tangletools
left a comment
Contributor
There was a problem hiding this comment.
✅ Auto-approved drewstone PR — e365da29
This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.
tangletools · auto-approval · reason: drewstone_author · 2026-07-01T02:29:01Z
This was referenced Jul 1, 2026
Merged
drewstone
added a commit
that referenced
this pull request
Jul 1, 2026
…oven, simplified (#430) Clears the remaining backlog from the examples DX audit (#428). Each fix verified (tc 0, lint clean; the offline-runnable ones run end-to-end $0). - intelligence-drop-in: the headline (OFF tier = $0 intelligence spend) was ASSERTED, never shown. Now the example records inference-only at OFF, flushes, and reads the exported span BACK off the collector to PROVE intelligence_usd=0 (throws otherwise). Dropped the confusing double-client ping ceremony; README rewritten to match (glosses Mode 0 / inference-vs-intelligence, adds a 'Going live' section). Runs $0. - strategy-suite: the authored doubleCheck was a near-clone of built-in refine. Rewrote it as a policy the built-ins DON'T have — require TWO consecutive passes before done (a flake/luck guard) — the real point of defineStrategy. Fixed the README (it listed 4 built-ins but the run compares 2). Runs $0. - agents-of-all-shapes: shipToTangleOtlp hand-rolled the OTLP resourceSpans envelope — the exact wire createOtelExporter produces (the primitive the example itself teaches). Routed it through createOtelExporter (~45 lines → ~25). Aligned npx→pnpm. Runs $0. - self-improving-loop: the n=3 'small-n mirage' warning was restated 4×; trimmed to one authoritative block (the⚠️ at the gate) + brief pointers. - agentic-data-creation: softened an unverifiable future-dated citation (arXiv 2606.25996 / 'Meta FAIR' / 'the paper's Table 1' with specific figures) stated as empirical fact → the real technique name (self-instruct, Wang et al. 2022) + 'illustrative target' framing for the by-construction numbers.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
From a 24-example DX audit (parallel read-only reviewers scored docs / code-simplicity / DX, flagged issues + a recommendation each). Result: 4 exemplary, 17 good, 3 needs-work, 0 broken, 0 redundant — the example set is healthy. This PR lands the highest-leverage, lowest-risk fixes: documentation truth.
$0CALIBRATE=1check.agent-knowledge"ships innode_modulesalready" (it does not — it's an optional peer you install; the example is excluded from CI typecheck), and it wrongly called itself "the primary, smallestrunLoopexample" (that'sdriver-loop).self-improving-coder(17b),intelligence-webcode(20b),webcode-matrix(9c).Backlog (code simplifications — next, if you want them)
$0-clamp by reading the span back (headline currently asserted, not shown)run-bridge/run-sandboxinto oneBACKENDknob (matches its own thesis)doubleCheckduplicates built-inrefine— make it show a policy the built-ins can'tDocs-only; no code touched.