docs(agents): add a verification fidelity ladder - #70
Merged
Conversation
Tuned for a repo whose failures are mostly failures to notice that nothing happened: rung 3 is running the task against a real project, rung 4 is proving the deployed app answers rather than trusting the deploy tool, and rung 6 is compiling from the packed tarball because Hex omits root dotfiles. Insider (GenericJam/insider) gates agent work on a named fidelity ladder and a rule that a lower rung may not stand in for a higher one. The principle was already here in pieces — "verify effects, not exit codes" in the agentic guide, the fast/slow split in .githooks/pre-push — but there was no named ladder an agent could point at, and so no vocabulary for "I verified at rung 3, rung 4 was unavailable." Every rung cites something that actually got through the rung below it, so the list reads as history rather than process. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Uh oh!
There was an error while loading. Please reload this page.
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 freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Tuned for a repo whose failures are mostly failures to notice that nothing
happened.
Rung 3 is running the task against a real project, because a task can unit-test
green and still skip its real work. Rung 4 is proving the deployed app answers
rather than trusting the deploy tool. Rung 6 is compiling from the packed
tarball, because Hex omits repository-root dotfiles and two releases have
already shipped broken that way.
Why
Borrowed from GenericJam/insider, which
gates agent work on a named fidelity ladder plus one rule: a lower rung may not
stand in for a higher one.
The principle was already here in pieces — "verify effects, not exit codes" in
the agentic guide, the deliberate fast/slow split in
.githooks/pre-push— butthere was no named ladder an agent could point at, and therefore no vocabulary
for the sentence that matters: "I verified at rung 3; rung 4 was unavailable
because no device was attached." Today an agent that runs
mix testand reportssuccess is not lying, and that is the problem.
Every rung cites something that actually got past the rung below it, so the list
reads as history rather than process.
Evidence
Docs only — no code paths touched.
.githooks/pre-pushran clean on push(format, Credo
--strictwith ex_slop, compile--warnings-as-errors).Not included
Deliberately left for a separate discussion rather than bundled here:
pre-commitwith insider'scheck-test-pairingco-change gateCODEOWNERSoverAGENTS.md,.githooks/anddecisions/commit-msgConventional Commits hook — not recommended aswritten: its regex rejects 30 of the last 40 subjects on
master, includingevery
Bump to x.y.zrelease commit and theMOB-nnn:style