feat(fleet): generate the coordination-strategy ladder as a drift-checked catalog - #261
Merged
Conversation
…cked catalog Skulk's fleet evasions don't fit the per-session evasion catalog: a fleet is a *shape* across N sessions, graded at the COORDINATION axis (benign/candidate/ fleet/campaign), and the tells that catch it are cross-session BINDINGS (fp_collision, trace_collision, template_similarity, shared_ticket, proxy_egress, arrival_regularity) computed in the coordination scorer — not per-session registry rules. Forcing them into the human/suspicious/bot table was misleading. New fleet-native generator (skulk/catalog.py, the analog of evasion_catalog): renders every registered strategy → the attacker class it models → the blue binding that catches it → its verdict tier, between GENERATED markers in fleet/README.md. Non-drifting: a completeness check fails if a new strategy has no catalog row, and a cross-guard asserts every exact-match binding matches what grade.assess() computes on the live members, so the catalog can't contradict the code. Wired task coordination-catalog + a CI --check (fleet job) + docs/docs:check. Readability: the README ladder is now a short primer (the two axes + the corroboration gate), the generated table sorted strongest→weakest conviction with a tier legend, and one frontier callout — instead of an 18-row wall of prose. fleet suite 36 passed, 97.86% coverage; ruff/mypy/headers clean.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Why
Skulk's fleet evasions never fit the
/evasionsper-session catalog. A per-session evader gets ahuman/suspicious/botverdict from registry rule-ids. A fleet is a shape emitted across N sessions, graded at the coordination axis (benign → candidate → fleet/campaign), and the tells that catch it are cross-session bindings (fp_collision,trace_collision,template_similarity,shared_ticket,proxy_egress,arrival_regularity) computed in the coordination scorer — not inregistry.yaml. Jamming them into the human/bot table was actively misleading, and the hand-authored ladder infleet/README.mdhad drifted into an 18-row wall of prose.What
A fleet-native generator —
skulk/catalog.py, the analog ofevasion_catalog.py— renders every registered strategy → the attacker class it models → the blue binding that catches it → its verdict tier, between<!-- GENERATED:coordination -->markers infleet/README.md.Non-drifting, and it can't lie about the code:
fleet-tier) binding matches whatgrade.assess()actually computes on the live members.Wired
task coordination-catalog+ a CI--checkin the fleet job + hooked intodocs/docs:check.Readability — the README ladder is now: a short primer (the two axes + the corroboration gate) → the generated table, sorted strongest→weakest conviction with a tier legend → one frontier callout. No more per-row wall.
The rendered table (18 strategies)
9
fleet(cluster axis) · 4campaign(population axis) · 4candidate(external-data-bound frontier) · 1below-candidate. Sorted by conviction strength;candidate/frontier rows grouped at the bottom under the "economic wall" note.Accuracy guarantees
fleet-tier rows are code-locked by the cross-guard againstgrade.assess().campaign-tier rows carry the README's grounded-live provenance (task coordination-live).harness→fleetscorer test — that would violate the "components never import each other" rule; the campaign tiers are already grounded end-to-end and by the harness's own coordination scenarios.Verification
fleetsuite 36 passed, 97.86% coverage; ruff/mypy/headers clean;python -m skulk.catalog --checkgreen.Follow-up (not in this PR)
Surface it on the site as
/coordination, with each binding row linking to the blue signal — the matched red/blue pair (bigger, detector-side).