Select operating points with a 2% disturbance margin - #187
Open
mjaskelioff wants to merge 5 commits into
Open
Conversation
Eighth analysed property: iid per-send loss, re-read per message. Per model: sweep_m?_pfail.py (exact per-message single-defect law at mu_eff = mu + (1-mu)p_fail, loss-injected MC behind --mc) and properties/transmission_unreliability.md; cross-model synthesis in comparison.md section 7 (Bottom line -> section 8); validate.py check (8) pins the loss-injected floods to the reference simulator at p_fail = 0. Findings: the mu_eff identity holds per edge, but per-message accounting moves the budgets -- the muted-publisher term loses its factor H (M2: ~1.7% -> 31.9%; M5 x2.3; M3 0.5% -> 0.92%) while the unshrunk H slightly tightens the in-dominated models (M1, M4). At 1% loss M3 is the only model out of budget; one per-link retry (+1% bandwidth) repairs it and keeps the frontier lead at 17.8%. The backlog's 'M4's doubled edge redundancy should win this axis' is killed: bidirectionality is a final-hop degree effect the identity already prices (12.8 tries vs M3's 9.6, fewer than M5/M1/M2), plus a bulk-regime-only interior pair-retry (z = -1.7 at the elevated cells, invisible at the delta tail). Validation: validate.py PASS (checks 1-8); all five model self-tests PASS; per-script gates: p_fail = 0 anchors exact on 40/40 graphs with published costs reproduced within 0.02%; elevated-cell law-vs-MC |z| <= 1.7 for M1/M2/M3/M5 (M4 residual recorded as finding); attempted-sends closed form within 0.06%; doc links resolve. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Replace the δ-cheapest selection rule in the model comparison with margin-admissible selection: a configuration must hold P(bad) ≤ 10⁻⁴ over μ_eff ≤ μ + 0.016 (p_max ≥ 2%). Selected points move accordingly (M1 F 24→25, M2 RF 24→25, M3 (12,8)→(13,7), M4 RF 8→9; M5 unchanged), and all per-model property files and scripts are updated to the new operating points. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
mjaskelioff
requested review from
EzequielPostan,
dfirsov,
jesusdiazvico and
will-break-it
as code owners
August 20, 2026 19:21
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.
Summary
Selects each model's operating point with an explicit disturbance margin instead of the bare δ criterion, and rewrites the model comparison around that selection.
transmission_unreliability.mdwith the per-message reading ε_msg,sweep_*_pfail.pyscripts, and sharedvalidate.py. Long MC passes sit behind opt-in flags and are never run by CI.comparison.mdrewritten in ASD-STE100 Simplified Technical English: short sentences, active voice, one term per concept; all numbers, tables, and links unchanged.Selected points
M1: F = 25 · M2: RF = 25 · M3: (13, 7) · M4: RF = 9 · M5: (9, 8). The M3/M4/M5 laws are churn-validated at exactly these parameters.
Caveat
Pooled across churn rounds, M3's law reads slightly optimistic (Stouffer +2.41, mechanism unidentified, direction conservative). If it resolves to a real correction, M3's admissibility at (13, 7) is the first thing to re-read.