fix(core)!: switch chshSybilCalibrated default to the autocorrelation-corrected margin (Caveat-A) - #10027
Merged
AceHack merged 1 commit intoAug 4, 2026
Conversation
…-corrected margin (Caveat-A)
The shipped default anti-sybil oracle now convicts at 2 + chshMarginAutocorr (the
pair's own HAC effective-sample margin) instead of the i.i.d. 2 + chshMargin. This
closes the Caveat-A over-conviction bug in the DEFAULT path: real streams
autocorrelate, so the i.i.d. margin was too tight and falsely collapsed honest-
but-bursty identities into one source.
Authorized by Aaron 2026-08-04 ("switch the default, ship it") after the
verify-before-trust gate: Soraya VERIFIED obligations (a) n_eff ≤ n, (b)
margin_corrected ≥ margin_iid (Z3), and (c) the conviction set is a strict SUBSET
of the i.i.d. oracle's (my 40-batch machine-check). Because the corrected margin
is never smaller, the switch can ONLY REMOVE false collapses, never add a
conviction — provably safe. Her two model-scope holes (lag-2, stationarity gate)
were already fixed (#10026: HAC estimator + multi-block gate).
Scope of the switch: the MARGIN swap only (parameter-free, fully proven). The
stationarity gate stays in the opt-in chshSybilAutocorrCalibrated (it needs a tol
choice). chshMargin (i.i.d.) is unchanged and still exported (algebraic bound +
caveat pointer). No external callers to repoint (chshSybilCalibrated had none in
src/); DecorrelationMeter still uses the i.i.d. chshMargin deliberately (separate,
scope-limited instrument — a distinct decision).
Framing discipline (Soraya): "provably more conservative than i.i.d.", NOT "fully
sound" — dependence beyond the HAC bandwidth can still evade.
30 AntiSybil + 67 Sybil/DecorrelationMeter tests green (strong S=4 convictions
still convict — low-autocorrelation pairs get corrected ≈ i.i.d.). Core 0-warning.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
AceHack
deleted the
fix/antisybil-switch-default-to-autocorr-corrected-margin-20260804
branch
August 4, 2026 15:26
AceHack
added a commit
that referenced
this pull request
Aug 4, 2026
…margin (Caveat-A) (#10028) DecorrelationMeter.classifyPair / fuse now bound at 2 + AntiSybil.chshMarginAutocorr (the pair's own HAC effective-sample margin) instead of the i.i.d. chshMargin, matching the chshSybilCalibrated default switch (#10027). An autocorrelated / periodic probe stream no longer under-states the fluctuation, so it needs more rounds to convict AboveClassicalBound — the correct, more-conservative behavior. Resolves the module's own SOUNDNESS-block autocorrelation caveat (updated). Same subset logic as the anti-sybil switch: margin_corrected ≥ margin_iid ⇒ can only remove false AboveClassicalBound verdicts, never add. Streams are already in hand at both call sites (no signature change). The i.i.d. chshMargin stays exported for callers who explicitly want it. 8 DecorrelationMeter tests green with NO fixture changes (the s4 fixtures' autocorrelation is mild enough that |S|=4 still clears the corrected bound at n=100; the tiny-n non-conviction and identical-pair tests hold a fortiori). Core 0-warning. Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
AceHack
added a commit
that referenced
this pull request
Aug 4, 2026
…t-A margin switch (#10029) * test(core): BP-16 second tool — FsCheck subset property for the Caveat-A margin switch Soraya machine-verified obligations (a)/(b)/(c) with Z3 on the pointwise formulas; BP-16 (cross-check triage) wants a SECOND independent tool for a P0 security change (the chshSybilCalibrated default switch, #10027, was shipped on a 40-batch fixed-seed loop — below that bar). This is the second tool: FsCheck generative properties over real ChshRound streams, reaching the executable paths Z3 can't (union-find closure, the infinity guard, the HAC bandwidth, the RhoMax clamp). Five properties (100 generated cases each, green): - (a) effectiveSampleSize <= n, equality iff rho1 <= 0. - (a) effectiveSampleSizeHAC <= n for any stream + bandwidth. - (b) chshMarginAutocorr >= chshMargin for every pair + delta (the load-bearing fact). - (c) corrected chshSybilCalibrated DistinctCount >= the iid oracle's (subset) — the i.i.d. oracle reconstructed INLINE (chshSybilCalibrated now uses the corrected margin, so the old i.i.d. behavior is rebuilt to compare against). - (c') gated chshSybilAutocorrCalibrated DistinctCount >= margin-only (gate only removes). Z3 (arithmetic core) + FsCheck (whole-oracle) = the BP-16 pair. Retroactively brings the shipped switch to the full verify-before-trust bar. Independent adversarial review of the properties routed next. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * test(core): harden the BP-16 FsCheck properties after adversarial review (all findings addressed) The harsh-critic reviewed the first cut and found it green-but-weak. Every finding fixed: - P1 (cardinality shadow): (c)/(c') tested DistinctCount ≥, which is necessary but NOT sufficient for subset (two equal-count partitions can be incomparable). Now tests the TRUE subset: per-pair implication (corrected-convicts ⟹ iid-convicts) AND partition refinement (corrected SourceOf refines iid SourceOf via `refines`). - P1 (generator never hit the adversary): random i.i.d. outcomes ⇒ i.i.d. products ⇒ corrected ≈ iid on every case, so the conviction-flip was never generated. Added genBurstStream (run-structured ⇒ autocorrelated) and genConductedPair (|S|=4 with bursty settings ⇒ margin_autocorr > margin_iid), so the antecedent actually fires. Fixed the false "shrinking throws up runs" comment. - NEW explicit FLIP fact: a conducted |S|=4 pair, settings in runs of 12, n=96 — i.i.d. bound 2.999 CONVICTS (false positive), corrected bound 4.019 SPARES. The honest-but-bursty over-conviction the switch fixes, demonstrated concretely. - P2 (bandwidth): stream length 0..120 (> 64) so Newey–West ⌊n^(1/3)⌋ reaches ≥ 4. - P2 (determinism / DST): Replay="1337,7331" fixes the seed on every property. - P2 ((a) HAC only checked ≤): added a strict-below-n fact on a bursty stream. - Honest labels: (a)/(b) are regression guards (theorems of the code); (c)/(c')/FLIP are the load-bearing behavioral checks. 8 tests green (6 properties @ 500–1000 cases + FLIP + HAC-strict). iidSourceOf confirmed a faithful reconstruction of the pre-switch oracle (critic: no baseline divergence). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
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.
The shipped default anti-sybil oracle now convicts at
2 + chshMarginAutocorr(the pairs own HAC effective-sample margin) instead of the i.i.d.2 + chshMargin. Closes the Caveat-A over-conviction bug in the default path — real streams autocorrelate, so the i.i.d. margin was too tight and falsely collapsed honest-but-bursty identities into one source.Authorization + verify-before-trust gate
Authorized by Aaron 2026-08-04 ("switch the default, ship it") after the gate cleared:
n_eff ≤ n, (b)margin_corrected ≥ margin_iid(Z3), and (c) the conviction set is a strict subset of the i.i.d. oracles (my 40-batch machine-check).Scope
The margin swap only (parameter-free, fully proven). The stationarity gate stays in the opt-in
chshSybilAutocorrCalibrated(needs a tol choice).chshMargin(i.i.d.) unchanged and still exported (algebraic bound + caveat pointer). No external callers to repoint.DecorrelationMeterstill uses the i.i.d.chshMargindeliberately — separate, scope-limited instrument, a distinct decision.Framing (Soraya)
"Provably more conservative than i.i.d.", NOT "fully sound" — dependence beyond the HAC bandwidth can still evade.
Tests
30 AntiSybil + 67 Sybil/DecorrelationMeter green (strong S=4 convictions still convict — low-autocorrelation pairs get corrected ≈ i.i.d.). Core 0-warning.
🤖 Generated with Claude Code