Repository navigation
Cost Explorer ChatGPT-parity follow-ups: lane charts, chart extras, evidence cards, projection routes #1400
Description
Follow-ups from the Cost Explorer ChatGPT-parity pass (branch feat/cost-explorer-chatgpt-parity). Everything in priorities 1–5 of the parity brief landed; the items below are what remains of the two ChatGPT explorers' feature sets, each with its blocker or reason noted. Nothing here blocks the parity PR.
Site 1 (aa-task-cost-explorer) remainders
- Band-comparison lane charts — groups 3–6 (fastest per intelligence band, smartest per speed band, fastest per cost band, cheapest per speed target) render as lane charts on Site 1; the parity PR ships the tables. Add lane rendering over the same band computations.
- Main-scatter comparison lines — smoothed upper-tail expectation, smoothed upper-envelope frontier, and the OLS cost–score trend line as toggleable overlays on the cost × score chart (the exact frontier and fitted-trend math is documented in Site 1's Data & methods tab).
- Per-chart view settings — third-metric limit modes per chart (unset / at most / at least + thresholds, plus per-band limits), chart-label modes (key choices / all / none), and magnification (fit / 150% / 200%).
- Unpriced-variant detail buttons under the main chart (configurations with a score but no published cost, listed as buttons opening the detail card). Partially covered today by chart 6 (scores-only bars) and the configurations table.
- Per-band recommendation overrides ("Tune this band": per-band tradeoff overrides, absolute limits per band, band split/merge). The parity PR has global tuners only.
- Provisional-estimate points + include-estimates toggle — blocked on measurement provenance in the Model data pipeline: workbook-free append-only event log + agent interpretation #1248 event log, same deferral as PR feat(pm): Cost Explorer — tabbed AA + Model explorer with frontier charts, perspectives, and full feature parity #1364's guest-estimate markers. Site 1's
reviewed-data.jsonholds the estimate set. - Evidence & anecdote cards (Data & methods) — the notes/event log from Model data pipeline: workbook-free append-only event log + agent interpretation #1248 is the home for these; seed from Site 1's five cards when the notes surface exists.
- AA directory table with OpenRouter ranks + OpenCode Console ids, and the OpenRouter/OpenCode coverage-check table (188 listings) — the fleet feed carries neither dataset today; needs a collector source before the page can render them.
- Design references & historical snapshots section (Site 1's Data & methods) — historical artifacts; link the Drive/site originals rather than re-hosting.
Site 2 (drew-coding-model-explorer) remainders
- Projection pricing routes — Site 2 prices the projection per route (Direct/benchmark API, OpenCode Console, OpenCode Go allowance). The parity PR prices from the recorded event-log rates only (one rate set per model). Needs per-route price events in the log before the selector means anything.
- Cache-storage hours in the projection — no storage rate exists in the price event log; add the field to price events first.
- Gemini promo vs Jan-2027 rates toggle — no promo windows recorded in price events yet.
- Throughput / latency factor sliders for the projected time (Site 2's speed assumptions).
- Shortlist side-by-side compare (up to 6) for catalog models — the parity PR has row detail cards and the recommendations shortlist; a pinned compare tray for the Model Explorer is the remaining piece.
Data notes
- Site 1's linked datasets (
reviewed-data.json,data.csv, CAI snapshots, coverage/cost/score audits, the CAI AInventory workbook) sit behind the ChatGPT login gate; retrieval needs the signed-in session or Aster's Drive upload (asked for in Slack #alignment). The parity PR proceeded from the signed-in browser inventory; pull the datasets when they land and diff the CAI observations against ainventory'sbenchmark_resultsas a cross-check.
Acceptance criteria
- Each remainder item listed in the body is either implemented or explicitly deferred with a named follow-up issue: Site 1 band-comparison lane charts (groups 3-6), main-scatter comparison-line overlays, per-chart view settings, unpriced-variant detail buttons, and per-band recommendation overrides.
- Lane charts render over the same band computations as the shipped tables (no divergent math).
- Nothing in the parity PR (feat(pm): Cost Explorer — ChatGPT-site parity (CAI perspective, Pareto tables, recommendations, bands, projection) #1390) regresses: existing Cost Explorer pages/charts still render and their tests pass.
Activity
developing-today commentedon Oct 27, 2026
CODE-491
dezren39 commentedon Oct 27, 2026
One more Site 2 remainder for the record: standalone Terminal-Bench 4.0 / SciCode benchmark perspectives. Site 2 lets either benchmark drive its explorer directly. Here, Terminal-Bench survives as a CAI component score (observations table + detail cards) and SciCode is not carried by the fleet feed at all — a standalone-benchmark perspective needs feed data for both before it can be built.
dezren39 commentedon Oct 27, 2026
Intake review
Fleet monitor intake enrichment, 2026-10-27.
Liveness: still live — this issue IS the remainder list from the merged parity PR #1390; no merged PR since implements these remainders (checked merged-PR titles this run).
Notes: A bundle of independent follow-ups, not one change. Suggested approach: split per site/chart area into separate small PRs against the parity code from #1390, starting with the lane charts (same band math, rendering only — lowest risk). The comparison-line overlays need the frontier/trend math documented in Site 1's Data & methods tab.
Risks: scope creep — the per-chart view-settings item is a settings system in miniature; sequence it last or split it out.
dezren39 commentedon Nov 1, 2026
Fresh parity audit — 2026-10-31 (Drew: "the explorer is missing views and Controls the old ones had that were supposed to be ported")
Re-audited the CURRENT live /aa page (aggregator on the laptop, feed status ok, 278 rankings / 608 catalog / 27 CAI rows) feature-by-feature against the 2026-10-27 signed-in inventory of the two ChatGPT explorers (
~/workspace/ai-subscriptions/chatgpt-explorer-sites-inventory-2026-10-27.md). This issue remains the tracker; the table below is the current state, not the state at the parity PR.Structure note: the current page has TWO tabs (AA Explorer, Model Explorer) with the Site-1 views as in-page sections. Recommendations / Explore charts / Band comparisons / Pareto exist as sections; Data & methods does not exist at all (zero "methods" text on the page).
Acceptance for the port (per missing item)
Filed from the 2026-10-31 parity audit at Drew's direction. A queue task for this port is being enqueued against this issue.
16 remaining items
dezren39 commentedon Nov 3, 2026
agent-note:v1
state: triaged (coverage pass)
Coverage pass 2026-11-03: reviewed in the fleet-wide per-issue pass (drift / overlap / already-handled / stale / blockers / ordering). Verdict: still live; queued in repo-health-code-pilot as triage at priority 40 (priority/normal). If this is already handled at current main, the triage closes it with evidence; if it needs splitting, the triage files the split.
Found via: coverage pass 2026-11-03
dezren39 commentedon Nov 3, 2026
Handed off — task 7689 (triage, lane code-f)
Claimed by repo-worker-code (lane code-f); execution lane assigned
opencode. Handed off to the opencode claimant to triage #1400 (Cost Explorer ChatGPT-parity follow-ups). Issue verified OPEN with no held/user-owned labels at handoff.dezren39 commentedon Nov 3, 2026
🤖 Agent note · findings
Verdict: still live as an umbrella, and it should be split before more execution work. Fresh state: #1400 OPEN, effort/l, no held/user-owned labels. The first Site 1 checklist item is done: "Band-comparison lane charts" shipped in PR #3972 (
feat(aggregator): add Cost Explorer band lane charts, merged 2026-11-02T15:45:33Z — the issue checklist still shows it unchecked).Ordering for the remainder (unblocked first):
Routing recommendation: carve one bounded
small_fix/front-end packet per item in group 1 rather than one effort/l task for the umbrella; keep groups 3 and 4 attached to their blockers. No packet was enqueued from this triage run — the active task history around #1400 already contains earlier slice work (task 4908 → PR #3972), so a split decision should be made once, in a dedicated pass, instead of adding a duplicate live packet blind.dezren39 commentedon Nov 3, 2026
🤖 Agent note · done
triaged. Cost Explorer ChatGPT-parity follow-ups: lane charts, chart extras, evidence cards, projection routes #1400 stays open as the parity umbrella; ordering and blockers are in the findings note above (group 1 unblocked UI slices first; Model data pipeline: workbook-free append-only event log + agent interpretation #1248/collector-blocked items not started). Lane-charts item verified done via merged PR feat(aggregator): add Cost Explorer band lane charts #3972.