Skip to content

Recommend standard speed and report API-equivalent speed premiums - #10

Merged
kixelated merged 8 commits into
mainfrom
quest/m0/fast-mode
Nov 7, 2026
Merged

kixelated merged 8 commits into
mainfrom
quest/m0/fast-mode

Conversation

@kixelated

@kixelated kixelated commented Nov 7, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

Recommend standard speed for justified current Claude Code/Codex premium preferences, naming the effective value, owner and replacement. Honor Claude per-session opt-in, managed false, active-profile environment disable, model support and current project/local settings. Honor Codex independent fast/ultrafast gates and local requirements; unresolved profile/project/provider ownership stays advisory. Historical premiums never recommend a change against current standard speed.

Add fast-mode to the existing shared analysis/report union, human/details output and schema-version-1 JSON. Separate observed response-speed premiums, current-setting estimates, and unknown billing metadata. Reprice at the same model's standard API rate with recorded cache categories/lifetimes, context tiers and geography held constant. Explicitly label this rate-only API-price comparison: Claude cache-switch effects and latency value are excluded; no subscription/credit or summed savings claims.

Native Codex TokenUsageRecord has no billed service tier. Remove the old requested-turn-tier fallback; retain requested controls separately without claiming they were billed. Claude recorded response usage can establish observed speed and standard fallback. Missing historical usage/prices does not block the current-setting warning or become invented dollars.

Complete and remove the fast-mode quest and parent reference. No publication, transcript repair or host-cost audit.

Choices

  • Report one card per known current owning control; distinct project controls keep disjoint request subsets. Unresolved ownership stays advisory.
  • Use current provider documentation checked 2026-11-07, including Codex 0.134.0's migration to NAME.config.toml profiles and current independent gates. Legacy/named selection and project trust/overrides are advisory where current launch ownership is unresolved. Unrecorded session/CLI toggles never come from old sessions.
  • Keep present public API-price equivalents distinct from historical invoices. Apply current speed to recorded usage independently of TTL/window/idle recommendations; observed premiums remain separately visible even when current speed is standard.
  • Extend the existing pricing path for supported OpenAI fast/priority and ultrafast rates. Preserve unsupported model/context combinations as unknown. No new analyzer or renderer entrypoint, dependency or arbitrary cutoff.

Public API and wire impact

Add the fast-mode check variant and structural Profile.speed/speedDirectories and Request.requestedSpeed metadata to the existing schema-version-1 report. Native Codex billed tiers and orphan Claude attempt tiers are now unknown instead of inherited from request settings. No CLI flags or exit-policy change: applicable current premium settings exit 1, advisory only exits 0, failed scans exit 2.

Validation

  • nix develop --command quest gate just check: formatting, TypeScript, JS build, 138 synthetic tests under Bun 1.4.4 and Node 24, quest validation, and packed standalone Node/npx/Node-free bunx consumers pass.
  • Coverage includes disabling overrides and exact owners, independent gates and ultrafast, unresolved profiles/providers, current-standard versus old premium, unknown usage/prices, explicit windows, current-setting-only exit 1, read-only inputs, malformed-selected exit 2, human/details/JSON parity, and preserved cache categories/context tiers/geography.
  • Merged fresh origin/main through 46f48a1, including packaging/release preparation and subsequent planning. Preserved completed quest removals and removed the newly introduced reference to fast-mode; reran the expanded full check.
  • No GitHub CI/online-review waiting and no full-host audit: corrupt real files are not repaired or bypassed.

Caused by

  • kixelated/token-max#3 used requested Codex turn tiers as billed usage tiers. This PR retains requested controls separately and keeps unobserved billing unknown. The fast-mode check itself completes the planned m0 feature.

Models

Size: S
Implementer: Codex, gpt-6.1-sol, xhigh effort

Review

Rewritten by quest review after each round.

  • Reviewer: Claude Code, claude-opus-5-5, low effort
  • Session: 0a43d748-1198-43a3-931c-83987b18b715, round 3
  • Head: 695d816
  • Base: origin/main at 46f48a1
  • Score: 8/10, Nothing under src or test changed since the last review, which already confirmed the fixes, so no open findings remain.
  • Findings: none

(Written by GPT-6.1 Sol with xhigh)

kixelated and others added 3 commits November 7, 2026 11:37
Quest-Claim: 8c8ebc38d434a1f5
…estimates

Co-Authored-By: GPT-6.1 Sol <noreply@openai.com>
…sons

Co-Authored-By: GPT-6.1 Sol <noreply@openai.com>
@kixelated kixelated changed the title Fast-mode check Recommend standard speed and report API-equivalent speed premiums Nov 7, 2026
@kixelated

Copy link
Copy Markdown
Collaborator Author

Design change: Emit one speed check per current owning control with disjoint request subsets. Distinct known project settings should each receive their applicable warning or keep result; a standard project must not mask another known premium project. Unresolved profile/project ownership stays advisory. The comparison still holds other recorded settings constant and never sums independent savings.

Also fixed all four initial review findings, with synthetic regressions for unsupported Opus history, the verified plain GPT-5.6 alias, nearest Codex project ownership and accurate error-file attribution. Full checks pass with 134 tests under both runtimes. Restarting local review for the committed head.

(Written by GPT-6.1 Sol with xhigh)

@kixelated
kixelated marked this pull request as ready for review November 7, 2026 20:00
@kixelated
kixelated enabled auto-merge (squash) November 7, 2026 20:01
@kixelated
kixelated merged commit 3cb995a into main Nov 7, 2026
15 of 17 checks passed
@kixelated
kixelated deleted the quest/m0/fast-mode branch November 7, 2026 20:02
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant