Files
vyndr/scripts/teeth-certified-probability.js
T
builtbykev be8e16aca9 Detection becomes repair: the curve is fitted on one forecaster, frozen, and named
The last release detected the violation and then served the certified state
anyway. A validator that changes nothing is decoration, so `servable:false` is
now load-bearing: an artifact that fails its policy returns
ARTIFACT_POLICY_BLOCKED with no number, and every probability-derived claim goes
with it. The gate sits inside the resolution, not beside the flag that turns the
shadow on, so no environment variable can reach past it — a test asserts
`resolve` never reads process.env at all. Shadow and live consume the SAME
decision, differing only in which promotion stage they demand.

Era mismatch still resolves to VERSION_MISMATCH rather than the new state. "This
artifact belongs to a different forecaster" is more precise than "policy
blocked", and the existing state already says it exactly.

THE REPAIR. `currentEraSource` filters on model_version in the QUERY, taking the
era from config/modelVersion so the query, the artifact and the validator all
read one identity. Measured on the actual fitted set, not a second count:
6,069 current-era rows, 0 wrong-era.

The procedure was then certified on current-era rows ONLY — four walk-forward
folds, training strictly before each evaluation block, 0 future rows in train on
every fold. All four improve; pooled n=3,108 gives Brier 0.24701 -> 0.24323,
delta -0.00378, CI [-0.00619,-0.00147] excluding zero; ECE falls in every fold.
Mapping spread inside support is 0.001-0.018. The prior mixed-era certification
did not substitute for this.

Policy B selected. A (era-filtered 65/35) and B (all current-era) are
statistically indistinguishable, A-B = +0.0001 CI [-0.00029,+0.00048], but B has
the better ECE (0.0064 vs 0.0109) and the holdout existed to certify the
PROCEDURE — it is not permanently withheld from the artifact that ships.
withheld_from_fit is 0.

FROZEN. `mlb-hits-isotonic@2026-09-03`: 6,069 rows, training_cutoff 2026-09-01
(distinct from fit_as_of 2026-09-03 — the newest observation admitted is not the
eligibility bound), 12 knots, source_digest 25919c16…, knot_digest 5ae940ea…,
served_curve_digest c24a9dc5…, 8 curve steps, 924 bytes, committed as JSON.

The runtime no longer fits. It loads. A test greps the service for fitIsotonic,
fromLedger and loadRows and requires all three absent, because the old behaviour
meant a user's number could move with no version, no review and no rollback, and
a past Read could not be reconstructed because its curve no longer existed.
New settled outcomes are forward evidence now; they cannot touch this curve.

Independent reconstruction from the declared training contract alone — fresh
read, fresh digest, fresh fit — reproduces every digest and the curve byte for
byte. Calling the builder twice would only have proven the builder deterministic.

Promotion is a frozen source constant. A snapshot cannot promote, a settlement
cannot promote, a successful fit cannot promote, and dropping a file into the
artifacts directory promotes nothing. Stage is APPROVED_FOR_SHADOW; live is
explicitly false.

Two coverage holes found by their own teeth. The promotion guard could be
deleted with every test still green, because the promoted file naturally agrees
with itself — extracted as `acceptFile` and tested on the case `load()` cannot
reach. And `validate(null)` returned no `servable` field at all, which is falsy
at a call site and so would have read as correct while asserting nothing.

Shadow OFF. Live OFF. CALIBRATION_DEPLOYED []. No frontend change.
Suite 404/404, 5,634 passed, 4 skipped. Teeth 26/26 + 10/10 + 23/23.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CQJeAG8vcDoL5zkiaJyVb8
2026-09-03 01:06:18 -04:00

166 lines
9.0 KiB
JavaScript

#!/usr/bin/env node
'use strict';
/**
* teeth-certified-probability — the NEW ground in this tranche.
*
* Teeth 6-15 and 19-23 of the order are already landed independently by
* scripts/teeth-probability-contract.js (23/23) and are not re-asserted here;
* this runner covers runtime observability, artifact identity, point-in-time
* evidence, shadow harmlessness, and the activation ordering.
*
* Teeth 17, 18 and 24 target a LIVE SERVING path that does not exist yet.
* They are recorded UNREACHABLE rather than asserted weakly.
*/
const fs = require('fs');
const path = require('path');
const crypto = require('crypto');
const { execSync } = require('child_process');
const ROOT = path.join(__dirname, '..');
const sha = (f) => crypto.createHash('sha256').update(fs.readFileSync(f)).digest('hex');
const codeOf = (s) => s.replace(/\/\*[\s\S]*?\*\//g, '').replace(/^\s*\/\/.*$/gm, '');
const results = [];
function runSuite(file) {
try { execSync(`npx jest ${file} --silent --testTimeout=45000`, { cwd: ROOT, stdio: 'pipe', timeout: 300000 });
return true; } catch { return false; }
}
function injectionTooth(id, name, file, find, replace, suite) {
const full = path.join(ROOT, file);
const before = fs.readFileSync(full, 'utf8');
const beforeSha = sha(full);
let landed = false, detail = '';
try {
if (!before.includes(find)) {
results.push({ id, name, landed: false, detail: `ANCHOR NOT FOUND in ${file} — the injection would have silently no-opped` });
return;
}
fs.writeFileSync(full, before.replace(find, replace));
if (fs.readFileSync(full, 'utf8') === before) throw new Error('injection produced no change');
landed = runSuite(suite) === false;
detail = landed ? `defect installed -> ${suite} FAILED as required`
: `defect installed and ${suite} STILL PASSED — coverage hole`;
} catch (e) { detail = 'threw: ' + e.message; }
finally {
fs.writeFileSync(full, before);
const ok = sha(full) === beforeSha;
detail += ok ? ' | restored byte-identical' : ' | RESTORE MISMATCH';
if (!ok) landed = false;
}
results.push({ id, name, landed, detail });
}
function logicTooth(id, name, fn) {
let landed = false, detail = '';
try { const r = fn(); landed = r.caught === true; detail = r.detail || ''; }
catch (e) { detail = 'threw: ' + e.message; }
results.push({ id, name, landed, detail });
}
// ── 1 — runtime SHA observability actually exists and is used ─────────────
logicTooth(1, 'runtime SHA verification waits for a snapshot despite working runtime status', () => {
const src = codeOf(fs.readFileSync(path.join(ROOT, 'src/routes/internal.js'), 'utf8'));
const block = src.slice(src.indexOf("router.get('/snapshot/status'"), src.indexOf("router.get('/snapshot/status'") + 4000);
const hasSha = /runtime:\s*\{[\s\S]*?code_sha:\s*codeSha\(\)/.test(block);
const hasStart = /started_at:\s*PROCESS_STARTED_AT/.test(block);
// and it must resolve from the SAME provenance source, not git HEAD
const ret = codeOf(fs.readFileSync(path.join(ROOT, 'src/services/retentionService.js'), 'utf8'));
const sameResolver = /function codeSha\(\)\s*\{\s*return process\.env\.SOURCE_COMMIT/.test(ret);
return { caught: hasSha && hasStart && sameResolver,
detail: `status exposes runtime.code_sha=${hasSha} started_at=${hasStart}; same resolver as provenance=${sameResolver}` };
});
// ── 2 — the estimator must carry a reconstructable identity ──────────────
injectionTooth(2, 'runtime estimator lacks a reconstructable artifact identity',
'src/services/model/fitPolicy.js',
` if (!artifact.knot_digest || !artifact.served_curve_digest) violations.push(VIOLATION.NO_ARTIFACT_IDENTITY);`,
` if (false) violations.push(VIOLATION.NO_ARTIFACT_IDENTITY);`,
'tests/unit/artifactGovernance.test.js');
// ── 3 — the artifact must be tied to the certified model era ─────────────
injectionTooth(3, 'runtime artifact differs from the certified artifact',
'src/services/model/artifactRegistry.js',
` estimator_type: 'isotonic',
stage: promoted.stage,`,
` estimator_type: 'low_param',
stage: promoted.stage,`,
'tests/unit/artifactGovernance.test.js');
// ── 4 — point-in-time evidence ───────────────────────────────────────────
injectionTooth(4, 'dynamic refit uses future settlement evidence for an earlier Read',
'src/services/model/calibrationService.js',
` .lt('game_date', cutoff), // STRICTLY before — the whole point`,
` .lte('game_date', cutoff),`,
'tests/unit/probabilityContract.test.js');
// ── 5 — the shadow may not move served product ───────────────────────────
injectionTooth(5, 'shadow changes current user-facing output',
'src/services/retentionService.js',
` return {
...r,
probability_contract: {`,
` return {
...r,
p_win: res.served_probability != null ? res.served_probability : r.p_win,
probability_contract: {`,
'tests/unit/probabilityContractShadow.test.js');
// ── 16 — activation ordering ─────────────────────────────────────────────
logicTooth(16, 'live serving activates before the shadow has passed', () => {
const snap = codeOf(fs.readFileSync(path.join(ROOT, 'src/services/snapshotService.js'), 'utf8'));
const deployedEmpty = /CALIBRATION_DEPLOYED\s*=\s*Object\.freeze\(\[\s*\]\)/.test(snap);
// no live consumer may read served_probability yet
const srcFiles = execSync(`grep -rl "served_probability" ${ROOT}/src ${ROOT}/web/src 2>/dev/null || true`)
.toString().trim().split('\n').filter(Boolean).map((f) => f.replace(ROOT + '/', ''));
const allowed = ['src/services/model/probabilityContract.js',
'src/services/model/probabilityContractService.js', 'src/services/retentionService.js'];
const leaked = srcFiles.filter((f) => !allowed.includes(f));
return { caught: deployedEmpty && leaked.length === 0,
detail: `CALIBRATION_DEPLOYED empty=${deployedEmpty}; served_probability referenced outside the contract layer: ${leaked.join(', ') || 'none'}` };
});
// ── the shadow flag itself ───────────────────────────────────────────────
logicTooth(25, 'shadow flag parses loosely or defaults ON', () => {
const snap = fs.readFileSync(path.join(ROOT, 'src/services/snapshotService.js'), 'utf8');
const pcSrc = fs.readFileSync(path.join(ROOT, 'src/services/model/probabilityContract.js'), 'utf8');
const strict = pcSrc.includes("String(raw || '') === '1'")
&& snap.includes("probabilityContract').shadowState().shadow === 'ON'");
const scoped = snap.includes("&& sp === 'mlb'");
const statScoped = snap.includes("sport: 'mlb', stat: 'hits'");
return { caught: strict && scoped && statScoped,
detail: `strict '1' compare=${strict}; sport-scoped=${scoped}; stat-scoped=${statScoped}` };
});
// ── 9 — the fit policy must not drift silently ───────────────────────────
injectionTooth(9, 'current fit policy silently changed before measurement',
'src/services/model/fitPolicy.js',
` model_version: 'engine1@2026-08-07-fullwindow',
data_selection: Object.freeze({`,
` model_version: 'engine1@2026-07-20',
data_selection: Object.freeze({`,
'tests/unit/fitPolicy.test.js');
// ── 10 — a refit may not become servable just because it ran ─────────────
injectionTooth(10, 'future refit becomes servable without a validity gate',
'src/services/model/fitPolicy.js',
` /** A policy-invalid artifact is NEVER servable. There is no override. */
servable: violations.length === 0,`,
` servable: true,`,
'tests/unit/fitPolicy.test.js');
// ── 10b — support may not be widened by a refit ──────────────────────────
injectionTooth(26, 'a refit widens the region it is trusted in',
'src/services/model/fitPolicy.js',
` violations.push(VIOLATION.SUPPORT_WIDENED);`,
` { /* widening allowed */ }`,
'tests/unit/fitPolicy.test.js');
const unreachable = [
{ id: 17, name: 'UI displays raw exact confidence for UNCERTIFIED', why: 'no live-serving UI path exists; nothing consumes served_probability yet' },
{ id: 18, name: 'live EV/Kelly/VALUE consumers bypass served_probability', why: 'live consumers are unchanged by design in this tranche — tooth 16 asserts none exist' },
{ id: 24, name: 'rollback rewrites historical probability contracts', why: 'nothing is activated, so there is no activation to roll back' },
];
const landed = results.filter((r) => r.landed).length;
console.log(JSON.stringify({ teeth_landed: `${landed}/${results.length}`, results, unreachable }, null, 2));
process.exit(landed === results.length ? 0 : 1);