DEM-PRODUCER-PLAN (reactor): the planned R:R becomes a verified provider fact - #89
Merged
Merged
Conversation
…act; goldens regenerated once, itemized DEM-GOV §9 DEM-PRODUCER-PLAN (owner-authorized 2026-08-25; afi-governance #55; §9 determinations D-2/D-4/D-5 in #56). Merge AFTER afi-core dem/plan-core and afi-config dem/plan-config. Source: - uss/cpjMapper.ts: the submitted plan survives as uss.plan (afi.trade-plan.v1, decimal strings, validated against the governed contract) or is refused at ingest (typed error → 422 trade_plan_invalid). TV/MarkitTick untouched. - enrichment/tradePlanVerification.ts (NEW): the D-DEM-5(6) law — every submitted level within [L − W, H + W] of the fetched window; plan geometry from prices only (D-DIR-3 by construction); R:R to the nearest target, 4 dp; refusal = TradePlanVerificationError (NodeConfigurationError). - providers/adapters/technicalLocalAdapter.ts: the producer — verifies the plan against ITS fetched candles and emits technical.plan (declared absences: no plan → no block; no stop/target → no R:R). - pipeline/nodes/laneView.ts: viewTechnical projects plan verbatim. - server.ts: 422 trade_plan_unverifiable (unwrapping NodeExecutionError.cause), 422 trade_plan_invalid; both persisted:false. - types/TradePlan.ts (NEW), types/UssLenses.ts, uss/ussValidator.ts: types. Registries/pins: the three fixture trees mirror afi-config (mapping 1.1.0, config, three implementationVersion moves, registration hash 5cb9b7a4…); registryLoader/officialArtifactPins pins rotated (pluginSetHash 36f911f4…). Goldens: the two plan-bearing CPJ fixtures re-authored inside the demo envelope (old levels were unverifiable by the law this slot lands); 12 goldens regenerated ONCE; every moved path itemized in INTENTIONAL_DIFFS.md (risk 0.9 → {0.5 via the provider's rr 1.4286 | 0.2 via the declared floor}; executionSummaryHash flips executed → degraded on the two goldens where the floor is the first fired default; identity-hash class on all 12). test/oracle/support/goldenDiff.mjs (NEW): the committed per-path differ. Tests: verification KATs (band, geometry, absences, error class), carrier (decimal strings, refusals), producer (role split, D-DIR-3 negative test, refusal, kernel floor, envelope-covers-fixtures), error-table rows for both 422s (no record), seam suites re-oracled to the composition reference with the mapping resolved by mappingRef from the fixture registry, golden scorerInput captured at the rubric call. 756/756; typecheck; compiled-build proofs (unavailable 503; Mongo persistence 10/10; oracle equivalence 7/7; SIGTERM) green locally. No scoring-law value moves. Sealed records untouched. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…n law An adversarial review of the branch (7 lenses, findings independently verified) found five real defects in the D-DEM-5(6) law. All five fixed here; no golden byte moves (every fix is refusal-side or message-side). - MAJOR: an entry RANGE was verified only at its conservative bound, so a range containing (or on the wrong side of) its own stop was ACCEPTED and an R:R emitted from one end. The stop must now clear the FAR bound as well: long -> stop < entryLow, short -> stop > entryHigh. Three refusal KATs (long straddle, short straddle, far bound EQUALS the stop) + the accept-side edge. - MINOR: the inclusive band edge was false on non-integer windows (binary float: 1.1 - 0.2 = 0.9000000000000001 refused an entry of exactly "0.9"). Relative tolerance (1e-12) at both edges + bounded-precision numbers in the refusal message; a non-integer-window KAT pins both edges. - MINOR: a non-array takeProfits (or a non-object levels) threw a bare TypeError, escaping the 422 unwrap. Both refuse properly now, with KATs. - MINOR: "half-up to 4 decimals" was not what Math.round(v*1e4)/1e4 does at binary ties. The law is documented as what it is (deterministic binary quantisation) and the unfavourable tie (0.00015 -> 0.0001) is pinned. - MINOR: missing edge KATs added: empty takeProfits array (no-target case), degenerate entry range (min === max), exotic numeric strings, an overflow-to-Infinity decimal string, malformed candles; the hedged short-geometry regex is tightened to its exact message. 756 -> 764 tests, all green; typecheck green; goldens byte-unchanged. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
Kilo Code Review could not run — your account is out of credits. Add credits or switch to a free model to enable reviews on this change. |
Contributor
Author
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
1 participant
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.
Wave 1 of 4 —
DEM-PRODUCER-PLAN, step 3 of 3. Merge after afi-core #36 and afi-config #91. This PR's CI is expected red until both are onmain(the reactor builds afi-coremainand reads afi-configmainthroughfile:links) — re-run it after those merges.How the gate is met, clause by clause
src/enrichment/tradePlanVerification.ts— every submitted level must lie within one observed-range width of the fetched window ([L−W, H+W]), the stop and targets must sit on opposite sides of the entry, and R:R is computed at the conservative entry bound. Any violation throwsTradePlanVerificationError(aNodeConfigurationError: no retry, pipeline abort) → 422trade_plan_unverifiable, no record. 31 KATs.tradePlanProducer.test.tsproves analyst-side inputs cannot move them and that the emitted facts equal the pure law applied to those two inputs alone.technical.planblock; a plan with no stop or no target → norrToFirstTarget(no R:R was claimed). Both are declared in the technical plugin's contract and enumerated in the mapping; the mapping's floor default fires as a recorded degradation.INTENTIONAL_DIFFS.md: resolved on plan-bearing CPJ submissions; residual on plan-less ones (see below).INTENTIONAL_DIFFS.mdsection itemizes every moved JSON path per golden, with an explicit byte-EQUAL list. A committed differ (test/oracle/support/goldenDiff.mjs) makes the audit reproducible — the gapDEM-BIND-PLANrecorded and nobody closed.Scored-value movement (authorized in class by D-DEM-7(2))
rrMultiplePlannedriskuwrScorecpj-blofin-perp-long.*The risk axis is no longer a single constant — it now spans {0.5, 0.2} across the corpus and is driven by the provider's plan where one exists.
One thing that needed a decision, and what I did
The two plan-bearing CPJ fixtures carried levels (BTC 42500/41800/43500) thousands of dollars outside any window the deterministic test feed can print — under the law this slot lands they would have been refused, and the golden corpus could not have been regenerated at all. I re-authored those two fixtures inside the feed's envelope (BTC 50000/49300/51000 → R:R 1.4286; ETH 3001.5, entry-only, keeping the non-integer decimal-string case) and added a test that the envelope contains every committed fixture plan, so the harness can never drift into silent refusal again. The consequential
ingestHash/inputHashmovement is itemized.Verification
764/764 green;
tsc --noEmit(strict, includestest/**) green; all four compiled-build proofs green against a real single-node replica set — honest-503 unavailable, 10 persistence checks, 7 oracle-equivalence checks, SIGTERM shutdown. Pre-open adversarial review (7 lenses, findings independently verified) found five real defects in the verification law; all five are fixed in the second commit.🤖 Generated with Claude Code