fix(pricing): resolve fable/mythos canary variants via fuzzy match by pitimon · Pull Request #188 · pitimon/TokenTracker

pitimon · 2026-08-22T14:44:09Z

PR Goal (one sentence)

Add curated fuzzy pricing rules so Claude Fable/Mythos canary model ids (e.g. claude-auto-pilot-fable-v1-canary) resolve to their pinned price instead of silently billing $0.

Scope

CLI (src/)
Dashboard (dashboard/)
Docs / CI / config

Checklist

npm run ci:local passes, or focused test commands are listed below with rationale
New dashboard/user-facing strings go through dashboard/src/content/copy.csv (N/A — no new strings)
Commits follow conventional style (feat: / fix: / refactor: / docs: / chore: / test: / ci:)
PR description explains why, not just what
Version bump files are in lockstep when release-bound (package.json, package-lock.json) — deferred to a follow-up publish PR per CLAUDE.md release workflow
npm publish state is recorded when release-bound — N/A for this PR, publish handled separately after merge

Why

src/lib/pricing/matcher.js resolves pricing through 7 tiers, and every tier (including the reverse-substring fuzzy fallback) requires the queue model id to contain the known price key as a contiguous substring. claude-auto-pilot-fable-v1-canary does not contain claude-fable-5 contiguously (-auto-pilot- interrupts it, and it ends in -v1-canary instead of -5), so it fell through every tier to miss → computeRowCost() returned 0 → the dashboards formatCost()returnsnullfor cost <= 0, hiding the$column entirely instead of showing$0.00`.

This is the same class of gap that #7 fixed for the literal claude-fable-5/claude-mythos-5 ids, but that fix does not cover vendor-renamed canary variants that wrap the name with extra segments.

Fixes#187.

Fix

Two curated fuzzy rules added to src/lib/pricing/curated-overrides.json, matching the existing kiro/hy3/composer fuzzy pattern:

{ "match": "fable", "ref": "claude-fable-5" },
{ "match": "mythos", "ref": "claude-mythos-5" },

Codex Context (required when requesting @codex review)

Delta since last Codex review: N/A, first review on this branch
Intended behavior / invariants: any model id containing the substring fable or mythos (case-insensitive) resolves via curated:fuzzy to the pinned Fable 5 / Mythos 5 price. Curated exact match and LiteLLM exact/dot match still take priority over this fuzzy rule per the existing resolve order.
Edge cases covered:claude-auto-pilot-fable-v1-canary, claude-mythos-v1-canary (regression test added); existing claude-fable-5/claude-mythos-5 exact ids continue to resolve via curated:exact (higher-priority tier, unaffected).
Tests run (command + result):
- node --test test/pricing.test.js → 34/34 pass (includes new regression case)
- NODE_ENV=test npm run ci:local → full gate pass (dashboard build, dashboard tests 318/318, product suite 1054+38, copy/UI-hardcode/architecture-guardrail/curated-expiry/outbound/openwiki validators)
Known gaps / out of scope: does not add a generic "strip auto-pilot/canary infix" normalizer — scoped to the two known-affected pinned names per issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187. A more general normalizer would be a separate proposal if more vendor-renamed variants surface.

Risk Layer Trigger (if any)

Public exposure / share links / unauthenticated access
Auth/session/token handling
Cross-endpoint invariants or shared logic
External gateway / environment constraints

(No risk-layer trigger — pure pricing-lookup data change, no I/O/auth/exposure surface touched.)

Regression Test Gate

Most likely regression surface

Existing fuzzy-matched models get overridden or shadowed by the new fable/mythos fuzzy rules if their ids happen to contain those substrings.

Verification method (choose at least one)

Full test/pricing.test.js suite re-run (34/34 pass, no existing case regressed) plus new regression assertions for the two canary ids from issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187.

Uncovered scope

Future vendor-renamed model families that are not literally named "fable" or "mythos" will need their own curated fuzzy rule when they surface (same as any other model addition).

claude-auto-pilot-fable-v1-canary and claude-mythos-*-canary model ids wrap the pinned claude-fable-5/claude-mythos-5 names with extra prefix/suffix segments (auto-pilot-, -v1-canary), so "fable"/"mythos" is not a contiguous substring of the curated exact key. Every non-fuzzy resolution tier in matcher.js requires a contiguous substring match, so these ids fell through to miss and billed $0 — the dashboard hid the cost column entirely (formatCost() returns null for cost <= 0). Add curated fuzzy rules for "fable" and "mythos", matching the existing kiro/hy3/composer fuzzy pattern. Fixes#187

chatgpt-codex-connector · 2026-08-22T14:44:17Z

To use Codex here, create a Codex account and connect to github.

Version bump for the fable/mythos canary pricing fix (#187, merged in #188). Updates the version-lockstep test's hardcoded package.json expectation in step. npm publish --dry-run verified: 174 files, tarball ok. Co-authored-by: itarun.p <itarun.p@somapait.com>

…closure (#194) Two follow-ups from the #187/#188 cross-verification review. 1. matcher.js: add a scoped denoise retry (issue #193). #188 fixed the fable/mythos canary regression with a per-name curated fuzzy rule, but left the structural gap open — any future auto-pilot/canary rename of an already-priced model would reproduce the same bug and require another one-line patch. The denoise retry strips the '-auto-pilot-' infix and trailing '-vN-canary'/'-canary' suffix and retries the full lookup ladder exactly once, only after every other tier has missed on the raw id, so a genuine exact match on the raw id (however unlikely) is never shadowed. Also narrows the #188 fuzzy rules from bare 'fable'/'mythos' substrings to 'claude-fable'/'claude-mythos'. Verified against the published pricing seed that a real collision risk existed: claude-mythos-preview is a genuine LiteLLM-priced model that happens to contain the substring 'mythos' — the bare rule would have silently mispriced any future model in the same shape that lacks its own exact LiteLLM/curated entry. 2. dashboard: wire the already-computed fuzzyPricingModels (from model-breakdown.ts) into ProviderBreakdownCard (issue #192). The provider card already discloses '{{count}} pricing missing' for unpriced models; it now also discloses '{{count}} pricing estimated' when any model resolved via substring/fuzzy match rather than an exact price, so a user sees when a total like $3,164.69 is a curated estimate, not a reconciled invoice figure. Verified: node --test test/pricing.test.js (39/39), dashboard ProviderBreakdownCard suite (10/10, incl. 2 new), npm run ci:local full gate (dashboard 320/320, product suite 1059/1059, all validators) — all green. Addresses #193, #192. Co-authored-by: itarun.p <itarun.p@somapait.com>

pitimon · 2026-08-22T14:44:09Z

PR Goal (one sentence)

Add curated fuzzy pricing rules so Claude Fable/Mythos canary model ids (e.g. claude-auto-pilot-fable-v1-canary) resolve to their pinned price instead of silently billing $0.

Scope

CLI (src/)
Dashboard (dashboard/)
Docs / CI / config

Checklist

npm run ci:local passes, or focused test commands are listed below with rationale
New dashboard/user-facing strings go through dashboard/src/content/copy.csv (N/A — no new strings)
Commits follow conventional style (feat: / fix: / refactor: / docs: / chore: / test: / ci:)
PR description explains why, not just what
Version bump files are in lockstep when release-bound (package.json, package-lock.json) — deferred to a follow-up publish PR per CLAUDE.md release workflow
npm publish state is recorded when release-bound — N/A for this PR, publish handled separately after merge

Why

src/lib/pricing/matcher.js resolves pricing through 7 tiers, and every tier (including the reverse-substring fuzzy fallback) requires the queue model id to contain the known price key as a contiguous substring. claude-auto-pilot-fable-v1-canary does not contain claude-fable-5 contiguously (-auto-pilot- interrupts it, and it ends in -v1-canary instead of -5), so it fell through every tier to miss → computeRowCost() returned 0 → the dashboards formatCost()returnsnullfor cost <= 0, hiding the$column entirely instead of showing$0.00`.

This is the same class of gap that #7 fixed for the literal claude-fable-5/claude-mythos-5 ids, but that fix does not cover vendor-renamed canary variants that wrap the name with extra segments.

Fixes#187.

Fix

Two curated fuzzy rules added to src/lib/pricing/curated-overrides.json, matching the existing kiro/hy3/composer fuzzy pattern:

{ "match": "fable", "ref": "claude-fable-5" },
{ "match": "mythos", "ref": "claude-mythos-5" },

Codex Context (required when requesting @codex review)

Delta since last Codex review: N/A, first review on this branch
Intended behavior / invariants: any model id containing the substring fable or mythos (case-insensitive) resolves via curated:fuzzy to the pinned Fable 5 / Mythos 5 price. Curated exact match and LiteLLM exact/dot match still take priority over this fuzzy rule per the existing resolve order.
Edge cases covered:claude-auto-pilot-fable-v1-canary, claude-mythos-v1-canary (regression test added); existing claude-fable-5/claude-mythos-5 exact ids continue to resolve via curated:exact (higher-priority tier, unaffected).
Tests run (command + result):
- node --test test/pricing.test.js → 34/34 pass (includes new regression case)
- NODE_ENV=test npm run ci:local → full gate pass (dashboard build, dashboard tests 318/318, product suite 1054+38, copy/UI-hardcode/architecture-guardrail/curated-expiry/outbound/openwiki validators)
Known gaps / out of scope: does not add a generic "strip auto-pilot/canary infix" normalizer — scoped to the two known-affected pinned names per issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187. A more general normalizer would be a separate proposal if more vendor-renamed variants surface.

Risk Layer Trigger (if any)

Public exposure / share links / unauthenticated access
Auth/session/token handling
Cross-endpoint invariants or shared logic
External gateway / environment constraints

(No risk-layer trigger — pure pricing-lookup data change, no I/O/auth/exposure surface touched.)

Regression Test Gate

Most likely regression surface

Existing fuzzy-matched models get overridden or shadowed by the new fable/mythos fuzzy rules if their ids happen to contain those substrings.

Verification method (choose at least one)

Full test/pricing.test.js suite re-run (34/34 pass, no existing case regressed) plus new regression assertions for the two canary ids from issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187.

Uncovered scope

Future vendor-renamed model families that are not literally named "fable" or "mythos" will need their own curated fuzzy rule when they surface (same as any other model addition).

claude-auto-pilot-fable-v1-canary and claude-mythos-*-canary model ids wrap the pinned claude-fable-5/claude-mythos-5 names with extra prefix/suffix segments (auto-pilot-, -v1-canary), so "fable"/"mythos" is not a contiguous substring of the curated exact key. Every non-fuzzy resolution tier in matcher.js requires a contiguous substring match, so these ids fell through to miss and billed $0 — the dashboard hid the cost column entirely (formatCost() returns null for cost <= 0). Add curated fuzzy rules for "fable" and "mythos", matching the existing kiro/hy3/composer fuzzy pattern. Fixes#187

chatgpt-codex-connector · 2026-08-22T14:44:17Z

To use Codex here, create a Codex account and connect to github.

Version bump for the fable/mythos canary pricing fix (#187, merged in #188). Updates the version-lockstep test's hardcoded package.json expectation in step. npm publish --dry-run verified: 174 files, tarball ok. Co-authored-by: itarun.p <itarun.p@somapait.com>

…closure (#194) Two follow-ups from the #187/#188 cross-verification review. 1. matcher.js: add a scoped denoise retry (issue #193). #188 fixed the fable/mythos canary regression with a per-name curated fuzzy rule, but left the structural gap open — any future auto-pilot/canary rename of an already-priced model would reproduce the same bug and require another one-line patch. The denoise retry strips the '-auto-pilot-' infix and trailing '-vN-canary'/'-canary' suffix and retries the full lookup ladder exactly once, only after every other tier has missed on the raw id, so a genuine exact match on the raw id (however unlikely) is never shadowed. Also narrows the #188 fuzzy rules from bare 'fable'/'mythos' substrings to 'claude-fable'/'claude-mythos'. Verified against the published pricing seed that a real collision risk existed: claude-mythos-preview is a genuine LiteLLM-priced model that happens to contain the substring 'mythos' — the bare rule would have silently mispriced any future model in the same shape that lacks its own exact LiteLLM/curated entry. 2. dashboard: wire the already-computed fuzzyPricingModels (from model-breakdown.ts) into ProviderBreakdownCard (issue #192). The provider card already discloses '{{count}} pricing missing' for unpriced models; it now also discloses '{{count}} pricing estimated' when any model resolved via substring/fuzzy match rather than an exact price, so a user sees when a total like $3,164.69 is a curated estimate, not a reconciled invoice figure. Verified: node --test test/pricing.test.js (39/39), dashboard ProviderBreakdownCard suite (10/10, incl. 2 new), npm run ci:local full gate (dashboard 320/320, product suite 1059/1059, all validators) — all green. Addresses #193, #192. Co-authored-by: itarun.p <itarun.p@somapait.com>

pitimon · 2026-08-22T14:44:09Z

PR Goal (one sentence)

Add curated fuzzy pricing rules so Claude Fable/Mythos canary model ids (e.g. claude-auto-pilot-fable-v1-canary) resolve to their pinned price instead of silently billing $0.

Scope

CLI (src/)
Dashboard (dashboard/)
Docs / CI / config

Checklist

npm run ci:local passes, or focused test commands are listed below with rationale
New dashboard/user-facing strings go through dashboard/src/content/copy.csv (N/A — no new strings)
Commits follow conventional style (feat: / fix: / refactor: / docs: / chore: / test: / ci:)
PR description explains why, not just what
Version bump files are in lockstep when release-bound (package.json, package-lock.json) — deferred to a follow-up publish PR per CLAUDE.md release workflow
npm publish state is recorded when release-bound — N/A for this PR, publish handled separately after merge

Why

src/lib/pricing/matcher.js resolves pricing through 7 tiers, and every tier (including the reverse-substring fuzzy fallback) requires the queue model id to contain the known price key as a contiguous substring. claude-auto-pilot-fable-v1-canary does not contain claude-fable-5 contiguously (-auto-pilot- interrupts it, and it ends in -v1-canary instead of -5), so it fell through every tier to miss → computeRowCost() returned 0 → the dashboards formatCost()returnsnullfor cost <= 0, hiding the$column entirely instead of showing$0.00`.

This is the same class of gap that #7 fixed for the literal claude-fable-5/claude-mythos-5 ids, but that fix does not cover vendor-renamed canary variants that wrap the name with extra segments.

Fixes#187.

Fix

Two curated fuzzy rules added to src/lib/pricing/curated-overrides.json, matching the existing kiro/hy3/composer fuzzy pattern:

{ "match": "fable", "ref": "claude-fable-5" },
{ "match": "mythos", "ref": "claude-mythos-5" },

Codex Context (required when requesting @codex review)

Delta since last Codex review: N/A, first review on this branch
Intended behavior / invariants: any model id containing the substring fable or mythos (case-insensitive) resolves via curated:fuzzy to the pinned Fable 5 / Mythos 5 price. Curated exact match and LiteLLM exact/dot match still take priority over this fuzzy rule per the existing resolve order.
Edge cases covered:claude-auto-pilot-fable-v1-canary, claude-mythos-v1-canary (regression test added); existing claude-fable-5/claude-mythos-5 exact ids continue to resolve via curated:exact (higher-priority tier, unaffected).
Tests run (command + result):
- node --test test/pricing.test.js → 34/34 pass (includes new regression case)
- NODE_ENV=test npm run ci:local → full gate pass (dashboard build, dashboard tests 318/318, product suite 1054+38, copy/UI-hardcode/architecture-guardrail/curated-expiry/outbound/openwiki validators)
Known gaps / out of scope: does not add a generic "strip auto-pilot/canary infix" normalizer — scoped to the two known-affected pinned names per issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187. A more general normalizer would be a separate proposal if more vendor-renamed variants surface.

Risk Layer Trigger (if any)

Public exposure / share links / unauthenticated access
Auth/session/token handling
Cross-endpoint invariants or shared logic
External gateway / environment constraints

(No risk-layer trigger — pure pricing-lookup data change, no I/O/auth/exposure surface touched.)

Regression Test Gate

Most likely regression surface

Existing fuzzy-matched models get overridden or shadowed by the new fable/mythos fuzzy rules if their ids happen to contain those substrings.

Verification method (choose at least one)

Full test/pricing.test.js suite re-run (34/34 pass, no existing case regressed) plus new regression assertions for the two canary ids from issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187.

Uncovered scope

Future vendor-renamed model families that are not literally named "fable" or "mythos" will need their own curated fuzzy rule when they surface (same as any other model addition).

claude-auto-pilot-fable-v1-canary and claude-mythos-*-canary model ids wrap the pinned claude-fable-5/claude-mythos-5 names with extra prefix/suffix segments (auto-pilot-, -v1-canary), so "fable"/"mythos" is not a contiguous substring of the curated exact key. Every non-fuzzy resolution tier in matcher.js requires a contiguous substring match, so these ids fell through to miss and billed $0 — the dashboard hid the cost column entirely (formatCost() returns null for cost <= 0). Add curated fuzzy rules for "fable" and "mythos", matching the existing kiro/hy3/composer fuzzy pattern. Fixes#187

chatgpt-codex-connector · 2026-08-22T14:44:17Z

To use Codex here, create a Codex account and connect to github.

Version bump for the fable/mythos canary pricing fix (#187, merged in #188). Updates the version-lockstep test's hardcoded package.json expectation in step. npm publish --dry-run verified: 174 files, tarball ok. Co-authored-by: itarun.p <itarun.p@somapait.com>

…closure (#194) Two follow-ups from the #187/#188 cross-verification review. 1. matcher.js: add a scoped denoise retry (issue #193). #188 fixed the fable/mythos canary regression with a per-name curated fuzzy rule, but left the structural gap open — any future auto-pilot/canary rename of an already-priced model would reproduce the same bug and require another one-line patch. The denoise retry strips the '-auto-pilot-' infix and trailing '-vN-canary'/'-canary' suffix and retries the full lookup ladder exactly once, only after every other tier has missed on the raw id, so a genuine exact match on the raw id (however unlikely) is never shadowed. Also narrows the #188 fuzzy rules from bare 'fable'/'mythos' substrings to 'claude-fable'/'claude-mythos'. Verified against the published pricing seed that a real collision risk existed: claude-mythos-preview is a genuine LiteLLM-priced model that happens to contain the substring 'mythos' — the bare rule would have silently mispriced any future model in the same shape that lacks its own exact LiteLLM/curated entry. 2. dashboard: wire the already-computed fuzzyPricingModels (from model-breakdown.ts) into ProviderBreakdownCard (issue #192). The provider card already discloses '{{count}} pricing missing' for unpriced models; it now also discloses '{{count}} pricing estimated' when any model resolved via substring/fuzzy match rather than an exact price, so a user sees when a total like $3,164.69 is a curated estimate, not a reconciled invoice figure. Verified: node --test test/pricing.test.js (39/39), dashboard ProviderBreakdownCard suite (10/10, incl. 2 new), npm run ci:local full gate (dashboard 320/320, product suite 1059/1059, all validators) — all green. Addresses #193, #192. Co-authored-by: itarun.p <itarun.p@somapait.com>

pitimon · 2026-08-22T14:44:09Z

PR Goal (one sentence)

Add curated fuzzy pricing rules so Claude Fable/Mythos canary model ids (e.g. claude-auto-pilot-fable-v1-canary) resolve to their pinned price instead of silently billing $0.

Scope

CLI (src/)
Dashboard (dashboard/)
Docs / CI / config

Checklist

npm run ci:local passes, or focused test commands are listed below with rationale
New dashboard/user-facing strings go through dashboard/src/content/copy.csv (N/A — no new strings)
Commits follow conventional style (feat: / fix: / refactor: / docs: / chore: / test: / ci:)
PR description explains why, not just what
Version bump files are in lockstep when release-bound (package.json, package-lock.json) — deferred to a follow-up publish PR per CLAUDE.md release workflow
npm publish state is recorded when release-bound — N/A for this PR, publish handled separately after merge

Why

src/lib/pricing/matcher.js resolves pricing through 7 tiers, and every tier (including the reverse-substring fuzzy fallback) requires the queue model id to contain the known price key as a contiguous substring. claude-auto-pilot-fable-v1-canary does not contain claude-fable-5 contiguously (-auto-pilot- interrupts it, and it ends in -v1-canary instead of -5), so it fell through every tier to miss → computeRowCost() returned 0 → the dashboards formatCost()returnsnullfor cost <= 0, hiding the$column entirely instead of showing$0.00`.

This is the same class of gap that #7 fixed for the literal claude-fable-5/claude-mythos-5 ids, but that fix does not cover vendor-renamed canary variants that wrap the name with extra segments.

Fixes#187.

Fix

Two curated fuzzy rules added to src/lib/pricing/curated-overrides.json, matching the existing kiro/hy3/composer fuzzy pattern:

{ "match": "fable", "ref": "claude-fable-5" },
{ "match": "mythos", "ref": "claude-mythos-5" },

Codex Context (required when requesting @codex review)

Delta since last Codex review: N/A, first review on this branch
Intended behavior / invariants: any model id containing the substring fable or mythos (case-insensitive) resolves via curated:fuzzy to the pinned Fable 5 / Mythos 5 price. Curated exact match and LiteLLM exact/dot match still take priority over this fuzzy rule per the existing resolve order.
Edge cases covered:claude-auto-pilot-fable-v1-canary, claude-mythos-v1-canary (regression test added); existing claude-fable-5/claude-mythos-5 exact ids continue to resolve via curated:exact (higher-priority tier, unaffected).
Tests run (command + result):
- node --test test/pricing.test.js → 34/34 pass (includes new regression case)
- NODE_ENV=test npm run ci:local → full gate pass (dashboard build, dashboard tests 318/318, product suite 1054+38, copy/UI-hardcode/architecture-guardrail/curated-expiry/outbound/openwiki validators)
Known gaps / out of scope: does not add a generic "strip auto-pilot/canary infix" normalizer — scoped to the two known-affected pinned names per issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187. A more general normalizer would be a separate proposal if more vendor-renamed variants surface.

Risk Layer Trigger (if any)

Public exposure / share links / unauthenticated access
Auth/session/token handling
Cross-endpoint invariants or shared logic
External gateway / environment constraints

(No risk-layer trigger — pure pricing-lookup data change, no I/O/auth/exposure surface touched.)

Regression Test Gate

Most likely regression surface

Existing fuzzy-matched models get overridden or shadowed by the new fable/mythos fuzzy rules if their ids happen to contain those substrings.

Verification method (choose at least one)

Full test/pricing.test.js suite re-run (34/34 pass, no existing case regressed) plus new regression assertions for the two canary ids from issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187.

Uncovered scope

Future vendor-renamed model families that are not literally named "fable" or "mythos" will need their own curated fuzzy rule when they surface (same as any other model addition).

claude-auto-pilot-fable-v1-canary and claude-mythos-*-canary model ids wrap the pinned claude-fable-5/claude-mythos-5 names with extra prefix/suffix segments (auto-pilot-, -v1-canary), so "fable"/"mythos" is not a contiguous substring of the curated exact key. Every non-fuzzy resolution tier in matcher.js requires a contiguous substring match, so these ids fell through to miss and billed $0 — the dashboard hid the cost column entirely (formatCost() returns null for cost <= 0). Add curated fuzzy rules for "fable" and "mythos", matching the existing kiro/hy3/composer fuzzy pattern. Fixes#187

chatgpt-codex-connector · 2026-08-22T14:44:17Z

To use Codex here, create a Codex account and connect to github.

Version bump for the fable/mythos canary pricing fix (#187, merged in #188). Updates the version-lockstep test's hardcoded package.json expectation in step. npm publish --dry-run verified: 174 files, tarball ok. Co-authored-by: itarun.p <itarun.p@somapait.com>

…closure (#194) Two follow-ups from the #187/#188 cross-verification review. 1. matcher.js: add a scoped denoise retry (issue #193). #188 fixed the fable/mythos canary regression with a per-name curated fuzzy rule, but left the structural gap open — any future auto-pilot/canary rename of an already-priced model would reproduce the same bug and require another one-line patch. The denoise retry strips the '-auto-pilot-' infix and trailing '-vN-canary'/'-canary' suffix and retries the full lookup ladder exactly once, only after every other tier has missed on the raw id, so a genuine exact match on the raw id (however unlikely) is never shadowed. Also narrows the #188 fuzzy rules from bare 'fable'/'mythos' substrings to 'claude-fable'/'claude-mythos'. Verified against the published pricing seed that a real collision risk existed: claude-mythos-preview is a genuine LiteLLM-priced model that happens to contain the substring 'mythos' — the bare rule would have silently mispriced any future model in the same shape that lacks its own exact LiteLLM/curated entry. 2. dashboard: wire the already-computed fuzzyPricingModels (from model-breakdown.ts) into ProviderBreakdownCard (issue #192). The provider card already discloses '{{count}} pricing missing' for unpriced models; it now also discloses '{{count}} pricing estimated' when any model resolved via substring/fuzzy match rather than an exact price, so a user sees when a total like $3,164.69 is a curated estimate, not a reconciled invoice figure. Verified: node --test test/pricing.test.js (39/39), dashboard ProviderBreakdownCard suite (10/10, incl. 2 new), npm run ci:local full gate (dashboard 320/320, product suite 1059/1059, all validators) — all green. Addresses #193, #192. Co-authored-by: itarun.p <itarun.p@somapait.com>

pitimon · 2026-08-22T14:44:09Z

PR Goal (one sentence)

Add curated fuzzy pricing rules so Claude Fable/Mythos canary model ids (e.g. claude-auto-pilot-fable-v1-canary) resolve to their pinned price instead of silently billing $0.

Scope

CLI (src/)
Dashboard (dashboard/)
Docs / CI / config

Checklist

npm run ci:local passes, or focused test commands are listed below with rationale
New dashboard/user-facing strings go through dashboard/src/content/copy.csv (N/A — no new strings)
Commits follow conventional style (feat: / fix: / refactor: / docs: / chore: / test: / ci:)
PR description explains why, not just what
Version bump files are in lockstep when release-bound (package.json, package-lock.json) — deferred to a follow-up publish PR per CLAUDE.md release workflow
npm publish state is recorded when release-bound — N/A for this PR, publish handled separately after merge

Why

src/lib/pricing/matcher.js resolves pricing through 7 tiers, and every tier (including the reverse-substring fuzzy fallback) requires the queue model id to contain the known price key as a contiguous substring. claude-auto-pilot-fable-v1-canary does not contain claude-fable-5 contiguously (-auto-pilot- interrupts it, and it ends in -v1-canary instead of -5), so it fell through every tier to miss → computeRowCost() returned 0 → the dashboards formatCost()returnsnullfor cost <= 0, hiding the$column entirely instead of showing$0.00`.

This is the same class of gap that #7 fixed for the literal claude-fable-5/claude-mythos-5 ids, but that fix does not cover vendor-renamed canary variants that wrap the name with extra segments.

Fixes#187.

Fix

Two curated fuzzy rules added to src/lib/pricing/curated-overrides.json, matching the existing kiro/hy3/composer fuzzy pattern:

{ "match": "fable", "ref": "claude-fable-5" },
{ "match": "mythos", "ref": "claude-mythos-5" },

Codex Context (required when requesting @codex review)

Delta since last Codex review: N/A, first review on this branch
Intended behavior / invariants: any model id containing the substring fable or mythos (case-insensitive) resolves via curated:fuzzy to the pinned Fable 5 / Mythos 5 price. Curated exact match and LiteLLM exact/dot match still take priority over this fuzzy rule per the existing resolve order.
Edge cases covered:claude-auto-pilot-fable-v1-canary, claude-mythos-v1-canary (regression test added); existing claude-fable-5/claude-mythos-5 exact ids continue to resolve via curated:exact (higher-priority tier, unaffected).
Tests run (command + result):
- node --test test/pricing.test.js → 34/34 pass (includes new regression case)
- NODE_ENV=test npm run ci:local → full gate pass (dashboard build, dashboard tests 318/318, product suite 1054+38, copy/UI-hardcode/architecture-guardrail/curated-expiry/outbound/openwiki validators)
Known gaps / out of scope: does not add a generic "strip auto-pilot/canary infix" normalizer — scoped to the two known-affected pinned names per issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187. A more general normalizer would be a separate proposal if more vendor-renamed variants surface.

Risk Layer Trigger (if any)

Public exposure / share links / unauthenticated access
Auth/session/token handling
Cross-endpoint invariants or shared logic
External gateway / environment constraints

(No risk-layer trigger — pure pricing-lookup data change, no I/O/auth/exposure surface touched.)

Regression Test Gate

Most likely regression surface

Existing fuzzy-matched models get overridden or shadowed by the new fable/mythos fuzzy rules if their ids happen to contain those substrings.

Verification method (choose at least one)

Full test/pricing.test.js suite re-run (34/34 pass, no existing case regressed) plus new regression assertions for the two canary ids from issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187.

Uncovered scope

Future vendor-renamed model families that are not literally named "fable" or "mythos" will need their own curated fuzzy rule when they surface (same as any other model addition).

claude-auto-pilot-fable-v1-canary and claude-mythos-*-canary model ids wrap the pinned claude-fable-5/claude-mythos-5 names with extra prefix/suffix segments (auto-pilot-, -v1-canary), so "fable"/"mythos" is not a contiguous substring of the curated exact key. Every non-fuzzy resolution tier in matcher.js requires a contiguous substring match, so these ids fell through to miss and billed $0 — the dashboard hid the cost column entirely (formatCost() returns null for cost <= 0). Add curated fuzzy rules for "fable" and "mythos", matching the existing kiro/hy3/composer fuzzy pattern. Fixes#187

chatgpt-codex-connector · 2026-08-22T14:44:17Z

To use Codex here, create a Codex account and connect to github.

Version bump for the fable/mythos canary pricing fix (#187, merged in #188). Updates the version-lockstep test's hardcoded package.json expectation in step. npm publish --dry-run verified: 174 files, tarball ok. Co-authored-by: itarun.p <itarun.p@somapait.com>

…closure (#194) Two follow-ups from the #187/#188 cross-verification review. 1. matcher.js: add a scoped denoise retry (issue #193). #188 fixed the fable/mythos canary regression with a per-name curated fuzzy rule, but left the structural gap open — any future auto-pilot/canary rename of an already-priced model would reproduce the same bug and require another one-line patch. The denoise retry strips the '-auto-pilot-' infix and trailing '-vN-canary'/'-canary' suffix and retries the full lookup ladder exactly once, only after every other tier has missed on the raw id, so a genuine exact match on the raw id (however unlikely) is never shadowed. Also narrows the #188 fuzzy rules from bare 'fable'/'mythos' substrings to 'claude-fable'/'claude-mythos'. Verified against the published pricing seed that a real collision risk existed: claude-mythos-preview is a genuine LiteLLM-priced model that happens to contain the substring 'mythos' — the bare rule would have silently mispriced any future model in the same shape that lacks its own exact LiteLLM/curated entry. 2. dashboard: wire the already-computed fuzzyPricingModels (from model-breakdown.ts) into ProviderBreakdownCard (issue #192). The provider card already discloses '{{count}} pricing missing' for unpriced models; it now also discloses '{{count}} pricing estimated' when any model resolved via substring/fuzzy match rather than an exact price, so a user sees when a total like $3,164.69 is a curated estimate, not a reconciled invoice figure. Verified: node --test test/pricing.test.js (39/39), dashboard ProviderBreakdownCard suite (10/10, incl. 2 new), npm run ci:local full gate (dashboard 320/320, product suite 1059/1059, all validators) — all green. Addresses #193, #192. Co-authored-by: itarun.p <itarun.p@somapait.com>

pitimon · 2026-08-22T14:44:09Z

PR Goal (one sentence)

Add curated fuzzy pricing rules so Claude Fable/Mythos canary model ids (e.g. claude-auto-pilot-fable-v1-canary) resolve to their pinned price instead of silently billing $0.

Scope

CLI (src/)
Dashboard (dashboard/)
Docs / CI / config

Checklist

npm run ci:local passes, or focused test commands are listed below with rationale
New dashboard/user-facing strings go through dashboard/src/content/copy.csv (N/A — no new strings)
Commits follow conventional style (feat: / fix: / refactor: / docs: / chore: / test: / ci:)
PR description explains why, not just what
Version bump files are in lockstep when release-bound (package.json, package-lock.json) — deferred to a follow-up publish PR per CLAUDE.md release workflow
npm publish state is recorded when release-bound — N/A for this PR, publish handled separately after merge

Why

src/lib/pricing/matcher.js resolves pricing through 7 tiers, and every tier (including the reverse-substring fuzzy fallback) requires the queue model id to contain the known price key as a contiguous substring. claude-auto-pilot-fable-v1-canary does not contain claude-fable-5 contiguously (-auto-pilot- interrupts it, and it ends in -v1-canary instead of -5), so it fell through every tier to miss → computeRowCost() returned 0 → the dashboards formatCost()returnsnullfor cost <= 0, hiding the$column entirely instead of showing$0.00`.

This is the same class of gap that #7 fixed for the literal claude-fable-5/claude-mythos-5 ids, but that fix does not cover vendor-renamed canary variants that wrap the name with extra segments.

Fixes#187.

Fix

Two curated fuzzy rules added to src/lib/pricing/curated-overrides.json, matching the existing kiro/hy3/composer fuzzy pattern:

{ "match": "fable", "ref": "claude-fable-5" },
{ "match": "mythos", "ref": "claude-mythos-5" },

Codex Context (required when requesting @codex review)

Delta since last Codex review: N/A, first review on this branch
Intended behavior / invariants: any model id containing the substring fable or mythos (case-insensitive) resolves via curated:fuzzy to the pinned Fable 5 / Mythos 5 price. Curated exact match and LiteLLM exact/dot match still take priority over this fuzzy rule per the existing resolve order.
Edge cases covered:claude-auto-pilot-fable-v1-canary, claude-mythos-v1-canary (regression test added); existing claude-fable-5/claude-mythos-5 exact ids continue to resolve via curated:exact (higher-priority tier, unaffected).
Tests run (command + result):
- node --test test/pricing.test.js → 34/34 pass (includes new regression case)
- NODE_ENV=test npm run ci:local → full gate pass (dashboard build, dashboard tests 318/318, product suite 1054+38, copy/UI-hardcode/architecture-guardrail/curated-expiry/outbound/openwiki validators)
Known gaps / out of scope: does not add a generic "strip auto-pilot/canary infix" normalizer — scoped to the two known-affected pinned names per issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187. A more general normalizer would be a separate proposal if more vendor-renamed variants surface.

Risk Layer Trigger (if any)

Public exposure / share links / unauthenticated access
Auth/session/token handling
Cross-endpoint invariants or shared logic
External gateway / environment constraints

(No risk-layer trigger — pure pricing-lookup data change, no I/O/auth/exposure surface touched.)

Regression Test Gate

Most likely regression surface

Existing fuzzy-matched models get overridden or shadowed by the new fable/mythos fuzzy rules if their ids happen to contain those substrings.

Verification method (choose at least one)

Full test/pricing.test.js suite re-run (34/34 pass, no existing case regressed) plus new regression assertions for the two canary ids from issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187.

Uncovered scope

Future vendor-renamed model families that are not literally named "fable" or "mythos" will need their own curated fuzzy rule when they surface (same as any other model addition).

claude-auto-pilot-fable-v1-canary and claude-mythos-*-canary model ids wrap the pinned claude-fable-5/claude-mythos-5 names with extra prefix/suffix segments (auto-pilot-, -v1-canary), so "fable"/"mythos" is not a contiguous substring of the curated exact key. Every non-fuzzy resolution tier in matcher.js requires a contiguous substring match, so these ids fell through to miss and billed $0 — the dashboard hid the cost column entirely (formatCost() returns null for cost <= 0). Add curated fuzzy rules for "fable" and "mythos", matching the existing kiro/hy3/composer fuzzy pattern. Fixes#187

chatgpt-codex-connector · 2026-08-22T14:44:17Z

To use Codex here, create a Codex account and connect to github.

Version bump for the fable/mythos canary pricing fix (#187, merged in #188). Updates the version-lockstep test's hardcoded package.json expectation in step. npm publish --dry-run verified: 174 files, tarball ok. Co-authored-by: itarun.p <itarun.p@somapait.com>

…closure (#194) Two follow-ups from the #187/#188 cross-verification review. 1. matcher.js: add a scoped denoise retry (issue #193). #188 fixed the fable/mythos canary regression with a per-name curated fuzzy rule, but left the structural gap open — any future auto-pilot/canary rename of an already-priced model would reproduce the same bug and require another one-line patch. The denoise retry strips the '-auto-pilot-' infix and trailing '-vN-canary'/'-canary' suffix and retries the full lookup ladder exactly once, only after every other tier has missed on the raw id, so a genuine exact match on the raw id (however unlikely) is never shadowed. Also narrows the #188 fuzzy rules from bare 'fable'/'mythos' substrings to 'claude-fable'/'claude-mythos'. Verified against the published pricing seed that a real collision risk existed: claude-mythos-preview is a genuine LiteLLM-priced model that happens to contain the substring 'mythos' — the bare rule would have silently mispriced any future model in the same shape that lacks its own exact LiteLLM/curated entry. 2. dashboard: wire the already-computed fuzzyPricingModels (from model-breakdown.ts) into ProviderBreakdownCard (issue #192). The provider card already discloses '{{count}} pricing missing' for unpriced models; it now also discloses '{{count}} pricing estimated' when any model resolved via substring/fuzzy match rather than an exact price, so a user sees when a total like $3,164.69 is a curated estimate, not a reconciled invoice figure. Verified: node --test test/pricing.test.js (39/39), dashboard ProviderBreakdownCard suite (10/10, incl. 2 new), npm run ci:local full gate (dashboard 320/320, product suite 1059/1059, all validators) — all green. Addresses #193, #192. Co-authored-by: itarun.p <itarun.p@somapait.com>

pitimon · 2026-08-22T14:44:09Z

PR Goal (one sentence)

Add curated fuzzy pricing rules so Claude Fable/Mythos canary model ids (e.g. claude-auto-pilot-fable-v1-canary) resolve to their pinned price instead of silently billing $0.

Scope

CLI (src/)
Dashboard (dashboard/)
Docs / CI / config

Checklist

npm run ci:local passes, or focused test commands are listed below with rationale
New dashboard/user-facing strings go through dashboard/src/content/copy.csv (N/A — no new strings)
Commits follow conventional style (feat: / fix: / refactor: / docs: / chore: / test: / ci:)
PR description explains why, not just what
Version bump files are in lockstep when release-bound (package.json, package-lock.json) — deferred to a follow-up publish PR per CLAUDE.md release workflow
npm publish state is recorded when release-bound — N/A for this PR, publish handled separately after merge

Why

src/lib/pricing/matcher.js resolves pricing through 7 tiers, and every tier (including the reverse-substring fuzzy fallback) requires the queue model id to contain the known price key as a contiguous substring. claude-auto-pilot-fable-v1-canary does not contain claude-fable-5 contiguously (-auto-pilot- interrupts it, and it ends in -v1-canary instead of -5), so it fell through every tier to miss → computeRowCost() returned 0 → the dashboards formatCost()returnsnullfor cost <= 0, hiding the$column entirely instead of showing$0.00`.

This is the same class of gap that #7 fixed for the literal claude-fable-5/claude-mythos-5 ids, but that fix does not cover vendor-renamed canary variants that wrap the name with extra segments.

Fixes#187.

Fix

Two curated fuzzy rules added to src/lib/pricing/curated-overrides.json, matching the existing kiro/hy3/composer fuzzy pattern:

{ "match": "fable", "ref": "claude-fable-5" },
{ "match": "mythos", "ref": "claude-mythos-5" },

Codex Context (required when requesting @codex review)

Delta since last Codex review: N/A, first review on this branch
Intended behavior / invariants: any model id containing the substring fable or mythos (case-insensitive) resolves via curated:fuzzy to the pinned Fable 5 / Mythos 5 price. Curated exact match and LiteLLM exact/dot match still take priority over this fuzzy rule per the existing resolve order.
Edge cases covered:claude-auto-pilot-fable-v1-canary, claude-mythos-v1-canary (regression test added); existing claude-fable-5/claude-mythos-5 exact ids continue to resolve via curated:exact (higher-priority tier, unaffected).
Tests run (command + result):
- node --test test/pricing.test.js → 34/34 pass (includes new regression case)
- NODE_ENV=test npm run ci:local → full gate pass (dashboard build, dashboard tests 318/318, product suite 1054+38, copy/UI-hardcode/architecture-guardrail/curated-expiry/outbound/openwiki validators)
Known gaps / out of scope: does not add a generic "strip auto-pilot/canary infix" normalizer — scoped to the two known-affected pinned names per issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187. A more general normalizer would be a separate proposal if more vendor-renamed variants surface.

Risk Layer Trigger (if any)

Public exposure / share links / unauthenticated access
Auth/session/token handling
Cross-endpoint invariants or shared logic
External gateway / environment constraints

(No risk-layer trigger — pure pricing-lookup data change, no I/O/auth/exposure surface touched.)

Regression Test Gate

Most likely regression surface

Existing fuzzy-matched models get overridden or shadowed by the new fable/mythos fuzzy rules if their ids happen to contain those substrings.

Verification method (choose at least one)

Full test/pricing.test.js suite re-run (34/34 pass, no existing case regressed) plus new regression assertions for the two canary ids from issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187.

Uncovered scope

Future vendor-renamed model families that are not literally named "fable" or "mythos" will need their own curated fuzzy rule when they surface (same as any other model addition).

claude-auto-pilot-fable-v1-canary and claude-mythos-*-canary model ids wrap the pinned claude-fable-5/claude-mythos-5 names with extra prefix/suffix segments (auto-pilot-, -v1-canary), so "fable"/"mythos" is not a contiguous substring of the curated exact key. Every non-fuzzy resolution tier in matcher.js requires a contiguous substring match, so these ids fell through to miss and billed $0 — the dashboard hid the cost column entirely (formatCost() returns null for cost <= 0). Add curated fuzzy rules for "fable" and "mythos", matching the existing kiro/hy3/composer fuzzy pattern. Fixes#187

chatgpt-codex-connector · 2026-08-22T14:44:17Z

To use Codex here, create a Codex account and connect to github.

Version bump for the fable/mythos canary pricing fix (#187, merged in #188). Updates the version-lockstep test's hardcoded package.json expectation in step. npm publish --dry-run verified: 174 files, tarball ok. Co-authored-by: itarun.p <itarun.p@somapait.com>

…closure (#194) Two follow-ups from the #187/#188 cross-verification review. 1. matcher.js: add a scoped denoise retry (issue #193). #188 fixed the fable/mythos canary regression with a per-name curated fuzzy rule, but left the structural gap open — any future auto-pilot/canary rename of an already-priced model would reproduce the same bug and require another one-line patch. The denoise retry strips the '-auto-pilot-' infix and trailing '-vN-canary'/'-canary' suffix and retries the full lookup ladder exactly once, only after every other tier has missed on the raw id, so a genuine exact match on the raw id (however unlikely) is never shadowed. Also narrows the #188 fuzzy rules from bare 'fable'/'mythos' substrings to 'claude-fable'/'claude-mythos'. Verified against the published pricing seed that a real collision risk existed: claude-mythos-preview is a genuine LiteLLM-priced model that happens to contain the substring 'mythos' — the bare rule would have silently mispriced any future model in the same shape that lacks its own exact LiteLLM/curated entry. 2. dashboard: wire the already-computed fuzzyPricingModels (from model-breakdown.ts) into ProviderBreakdownCard (issue #192). The provider card already discloses '{{count}} pricing missing' for unpriced models; it now also discloses '{{count}} pricing estimated' when any model resolved via substring/fuzzy match rather than an exact price, so a user sees when a total like $3,164.69 is a curated estimate, not a reconciled invoice figure. Verified: node --test test/pricing.test.js (39/39), dashboard ProviderBreakdownCard suite (10/10, incl. 2 new), npm run ci:local full gate (dashboard 320/320, product suite 1059/1059, all validators) — all green. Addresses #193, #192. Co-authored-by: itarun.p <itarun.p@somapait.com>

pitimon · 2026-08-22T14:44:09Z

PR Goal (one sentence)

Add curated fuzzy pricing rules so Claude Fable/Mythos canary model ids (e.g. claude-auto-pilot-fable-v1-canary) resolve to their pinned price instead of silently billing $0.

Scope

CLI (src/)
Dashboard (dashboard/)
Docs / CI / config

Checklist

npm run ci:local passes, or focused test commands are listed below with rationale
New dashboard/user-facing strings go through dashboard/src/content/copy.csv (N/A — no new strings)
Commits follow conventional style (feat: / fix: / refactor: / docs: / chore: / test: / ci:)
PR description explains why, not just what
Version bump files are in lockstep when release-bound (package.json, package-lock.json) — deferred to a follow-up publish PR per CLAUDE.md release workflow
npm publish state is recorded when release-bound — N/A for this PR, publish handled separately after merge

Why

src/lib/pricing/matcher.js resolves pricing through 7 tiers, and every tier (including the reverse-substring fuzzy fallback) requires the queue model id to contain the known price key as a contiguous substring. claude-auto-pilot-fable-v1-canary does not contain claude-fable-5 contiguously (-auto-pilot- interrupts it, and it ends in -v1-canary instead of -5), so it fell through every tier to miss → computeRowCost() returned 0 → the dashboards formatCost()returnsnullfor cost <= 0, hiding the$column entirely instead of showing$0.00`.

This is the same class of gap that #7 fixed for the literal claude-fable-5/claude-mythos-5 ids, but that fix does not cover vendor-renamed canary variants that wrap the name with extra segments.

Fixes#187.

Fix

Two curated fuzzy rules added to src/lib/pricing/curated-overrides.json, matching the existing kiro/hy3/composer fuzzy pattern:

{ "match": "fable", "ref": "claude-fable-5" },
{ "match": "mythos", "ref": "claude-mythos-5" },

Codex Context (required when requesting @codex review)

Delta since last Codex review: N/A, first review on this branch
Intended behavior / invariants: any model id containing the substring fable or mythos (case-insensitive) resolves via curated:fuzzy to the pinned Fable 5 / Mythos 5 price. Curated exact match and LiteLLM exact/dot match still take priority over this fuzzy rule per the existing resolve order.
Edge cases covered:claude-auto-pilot-fable-v1-canary, claude-mythos-v1-canary (regression test added); existing claude-fable-5/claude-mythos-5 exact ids continue to resolve via curated:exact (higher-priority tier, unaffected).
Tests run (command + result):
- node --test test/pricing.test.js → 34/34 pass (includes new regression case)
- NODE_ENV=test npm run ci:local → full gate pass (dashboard build, dashboard tests 318/318, product suite 1054+38, copy/UI-hardcode/architecture-guardrail/curated-expiry/outbound/openwiki validators)
Known gaps / out of scope: does not add a generic "strip auto-pilot/canary infix" normalizer — scoped to the two known-affected pinned names per issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187. A more general normalizer would be a separate proposal if more vendor-renamed variants surface.

Risk Layer Trigger (if any)

Public exposure / share links / unauthenticated access
Auth/session/token handling
Cross-endpoint invariants or shared logic
External gateway / environment constraints

(No risk-layer trigger — pure pricing-lookup data change, no I/O/auth/exposure surface touched.)

Regression Test Gate

Most likely regression surface

Existing fuzzy-matched models get overridden or shadowed by the new fable/mythos fuzzy rules if their ids happen to contain those substrings.

Verification method (choose at least one)

Full test/pricing.test.js suite re-run (34/34 pass, no existing case regressed) plus new regression assertions for the two canary ids from issue Claude auto-pilot/canary model ids (fable/mythos) bill $0 — fuzzy pricing match misses non-contiguous substrings #187.

Uncovered scope

Future vendor-renamed model families that are not literally named "fable" or "mythos" will need their own curated fuzzy rule when they surface (same as any other model addition).

claude-auto-pilot-fable-v1-canary and claude-mythos-*-canary model ids wrap the pinned claude-fable-5/claude-mythos-5 names with extra prefix/suffix segments (auto-pilot-, -v1-canary), so "fable"/"mythos" is not a contiguous substring of the curated exact key. Every non-fuzzy resolution tier in matcher.js requires a contiguous substring match, so these ids fell through to miss and billed $0 — the dashboard hid the cost column entirely (formatCost() returns null for cost <= 0). Add curated fuzzy rules for "fable" and "mythos", matching the existing kiro/hy3/composer fuzzy pattern. Fixes#187

chatgpt-codex-connector · 2026-08-22T14:44:17Z

To use Codex here, create a Codex account and connect to github.

Version bump for the fable/mythos canary pricing fix (#187, merged in #188). Updates the version-lockstep test's hardcoded package.json expectation in step. npm publish --dry-run verified: 174 files, tarball ok. Co-authored-by: itarun.p <itarun.p@somapait.com>

…closure (#194) Two follow-ups from the #187/#188 cross-verification review. 1. matcher.js: add a scoped denoise retry (issue #193). #188 fixed the fable/mythos canary regression with a per-name curated fuzzy rule, but left the structural gap open — any future auto-pilot/canary rename of an already-priced model would reproduce the same bug and require another one-line patch. The denoise retry strips the '-auto-pilot-' infix and trailing '-vN-canary'/'-canary' suffix and retries the full lookup ladder exactly once, only after every other tier has missed on the raw id, so a genuine exact match on the raw id (however unlikely) is never shadowed. Also narrows the #188 fuzzy rules from bare 'fable'/'mythos' substrings to 'claude-fable'/'claude-mythos'. Verified against the published pricing seed that a real collision risk existed: claude-mythos-preview is a genuine LiteLLM-priced model that happens to contain the substring 'mythos' — the bare rule would have silently mispriced any future model in the same shape that lacks its own exact LiteLLM/curated entry. 2. dashboard: wire the already-computed fuzzyPricingModels (from model-breakdown.ts) into ProviderBreakdownCard (issue #192). The provider card already discloses '{{count}} pricing missing' for unpriced models; it now also discloses '{{count}} pricing estimated' when any model resolved via substring/fuzzy match rather than an exact price, so a user sees when a total like $3,164.69 is a curated estimate, not a reconciled invoice figure. Verified: node --test test/pricing.test.js (39/39), dashboard ProviderBreakdownCard suite (10/10, incl. 2 new), npm run ci:local full gate (dashboard 320/320, product suite 1059/1059, all validators) — all green. Addresses #193, #192. Co-authored-by: itarun.p <itarun.p@somapait.com>