Total-usage charts, additional Copilot collection and official price references
Before repairs, inspect current source/lockfiles and permitted local fields. Results: repair record.
Behavior and compatibility
Current reference prices must cover the same retained range as usage. Select individual details or archived summaries per source partition, using identical daily/weekly/monthly archive rules. Hourly selection excludes whole-day totals. Archived summaries use saved components only: no zero-filled splits or per-call context tiers inferred from daily input. Multiple tiers without recoverable per-call tier show lower/upper costs and reasons; known-component intervals do not represent full cost. Current references never overwrite historical prices/archives. Retain integer-product precision, sum then round to avoid losing small costs. Model details merge by source provider/spelling key; reference supplier/tiers/ snapshots appear within that model row.
Dashboard rules:
- Managed telemetry roots go only to OTel. Shared manual roots honor confirmed file ownership; another adapter cannot collect an already assigned physical file. Reevaluate old wrong registrations without deleting usage/revisions/diagnostics. Incorrect registrations without usage history in that directory do not appear as installed agents. Actual unknown manual formats keep diagnostics; missing historical sources gray out, permission errors remain read failures.
- Keep original model name, spelling comparison key and verified API reference identity separate. Spaces/underscores/hyphens may normalize for comparison; retain version decimals/unknown suffixes. Dynamic profiles preserve effective dates. Original source/channel/historical prices independent; model rows explain missing rates/usage.
- Missing cost-curve points stay gaps, known zero stays zero. ECharts missing sentinel never enters money formatting.
- Sum Copilot input/output only with complete matching coverage. Native turn’s last-call input and whole-turn output provide a display-only observed-total lower bound, with explanation; never redefine it as complete total tokens or include in full-total statistics.
- Setting switches share visuals/focus/keyboard behavior; multiple choices retain checkbox semantics and clickable labels.
API reference cost card belongs in today/trend metric grid: eight columns wide, four medium, two/one narrow. Account quota remains separate. Distinguish missing priceable source tokens from no applicable price; partial estimates explain missing input splits/output/rate components. Unknown tokens remain unknown; known-zero usage may price to zero with an applicable rate.
User-requested CNY→USD display preserves original CNY amount/rate alongside approximate USD, without currency merging/history rewriting. Versioned ECB 2026-10-02 snapshot EUR/USD=1.1225, EUR/CNY=7.5259, CNY→USD=1.1225/7.5259; display date/source, no automatic network refresh. Unknown currencies/unsafe numbers unconverted. Amounts/rates converted and rounded in their respective integer units. ECB reference rates.
Codex automatic file reevaluation retains individual usage/revisions/comparison diagnostics. Cumulative mismatch stays visible without independently lowering read health. New individual records do not depend on snapshots; invalid legacy total/last fields still worsen health. Official TokenCountEvent.info may be absent/null: no usage/call/zero filling. Invalid rows, individual usage shape/time errors remain after later good batches. Register Kilo versions only with redacted per-version samples; empty sessions without assistant usage cannot verify format or a version range. Updated parser/version verification automatically rescans old cursors. Compatibility notices explain automatic readable-usage checks and review after support updates. Missing version verification and actual read errors stay distinct; notices cannot be cleared through unsupported version claims. Kilo uses each message’s session.version; highest database version never replaces another session’s basis. Mixed compatibility survives incremental rounds, reevaluated after rule-upgrade replay.
- Total view draws total tokens only per group; unknown gaps remain, input/output in their own charts/tooltips. Share chart switches total/input/output; agents without totals keep names and explicit unknown values.
- Call charts show agent names/counts, including hoverable points for a single period. Native Copilot usage observations are not calls; verified hyphen/decimal model spellings group together.
- Automatically discover locally configured Copilot file exports. Accept only verified CLIENT chat spans, including chat followed by model; skip metrics/logs/invoke_agent/noncall records. Calls without reported tokens still count; cache fields unknown, reasoning subset of output. Small overview text says VS Code/Agent Host/latest CLI/JetBrains exports supplement per-call observations. CodeBuddy CLI already has local sessions; export is optional supplementation.
- Without common call IDs, choose native session or OTel contributions for each original host/user/session/statistical local day. Observed Copilot OTel replaces matching native contribution; other dates/sessions retain native data. No deduplication by equal time/tokens. Enabling-day exports may cover only later calls; show limited coverage. Preserve native records/revision history; later native replay cannot add them again. Missing session identity cannot replace native records. Same-source file/HTTP copies deduplicate by trace+span and original host/user. Equal span IDs in different traces remain different calls. Replay old span-only cursors, retaining old records/removing old contribution without clearing the DB.
- Costs off by default. Exact provider/model/user channel prices first. Without exact rows, missing provider/channel can still use verified official same-model API reference rates. Without channel preference choose official global, preferring api labels but also official provider-name channel labels. Otherwise choose official channel with definite currency; remaining channel/currency ambiguity stays unpriced. Preserve real provider/currency/ fallback_event_count. Never price another model from the same series; subscription-only models without API rates stay unknown. Official Kimi model table maps k3/k3-256k to K3 for kimi-k3 reference only. Alias applies to bare model_raw k3/k3-256k from native Kilo/oh-my-pi custom Kimi providers and prefixed forms, official moonshot assignment. Keep original profile; exact profile-channel rates first. Strip custom namespaces only when prefix matches the record provider or verified prefix table. Verified backup profiles follow the same rule; unknown namespaces/suffixes retained. Official GPT-5.5 snapshot gpt-5.5-2026-04-23 may reference gpt-5.5; no inferred other dates. Official Kimi release notes map kimi-for-coding from 2026-09-11 to kimi-k2.8-preview, without changing earlier identity. No verified original-provider public API rate yet; current reference follows the explicit exception below, occurrence-time estimate remains missing. Tencent/other-channel prices are not labeled Moonshot official rates. Native official CodeBuddy catalog identifies hy4-preview-f as Hy4 preview, using Tencent official CNY reference. Centralize alias/spelling rules; exact catalog ID before spelling normalization/context tier, conflicts unpriced. All tiers from one preferred snapshot, without mixing new base/old long rates. User confirmed k28-agent-preview as Kimi K2.8 Preview on 2026-10-05 and authorized K2.7 when official prices absent. Identity stays kimi-k2.8-preview. Current API reference uses official kimi-k2.7-code only without exact channel/same-model official rate. This listed cross-model exception does not apply to similar models or channel/tier ambiguity. Amounts/ summaries/rate details identify substitute model. Occurrence-time estimates never use it; no historical amount rewriting.
- Overview/trend show API references/partial coverage/official fallback/unpriced reasons. Known usage observations can receive partial estimates; priced-event count is not call count. Background prices never rewrite occurrence-time estimates. Rule repairs recalculate retained detail on unarchived days once.
- Trend defaults: calls/usage side by side, costs full width, weekday/two share charts on one evenly divided row. User adjustments/old layouts retained; grid fills empty columns, narrow screens one column. Seven range metrics plus quota form eight cards, wide two rows/four columns, narrow two/one, without blank space beside quota.
Official provider references
Page text checked 2026-10-02. Model series narrow candidate official suppliers only; exact model must still have a price row, without invented series-based rates.
| Series | Official supplier/catalog ID | Source |
|---|---|---|
| GPT/OpenAI o | OpenAI/openai | OpenAI pricing |
| Claude | Anthropic/anthropic | Claude pricing |
| Gemini | Google/google | Gemini pricing |
| Kimi/Moonshot | Moonshot AI/moonshot/moonshotai, including CN | Kimi pricing |
| GLM | Zhipu/zhipuai; international Z.ai/zai | Z.ai pricing |
| DeepSeek | DeepSeek/deepseek | Official pricing, public HTTP 200/23,982 bytes retained under ignored build/dashboard-repair/ |
Kimi profiles: Kimi Code model table; dates: release notes. HY4 uses Tencent official prices, Guangzhou standard API CNY/million input 6/output 18/cache hit 0.3. 2026-10-03 snapshot effective from verification date; free CodeBuddy access does not establish free API pricing. Local details/results: second feedback record. 2026-10-02 snapshot adds Claude Opus 4.8 and GPT-4o mini official 2024-07-18 snapshot. Standard USD/million: Opus input 5/output 25/cache hit 0.50/write 5m 6.25/write 1h 10; GPT-4o mini input 0.15/ cache hit 0.075/output 0.60. Effective from verification date, without invented historical dates or zero prices for absent cache components.
Copilot file shape: archived fixed fileExporters source and 30 native CLIENT chat spans. CLI individual/aggregate distinction: official OTel reference. Native VS Code acceptance does not verify new CLI/JetBrains versions. Input/cache inclusion and reasoning output subset checked against OTel GenAI semantics. SDK CLIENT=2 differs from OTLP CLIENT=3; receiver converts to text enums to avoid mixing them.
Failures, validation and rollback
Missing/damaged exports show existing observable coverage; no model calls to manufacture samples. Original local files read-only; temporary validation databases/redacted output under root build/dashboard-repair/. Regressions cover old cursors/rescans/read order, host/user isolation, enabling-day range, hover, official reference with unknown channel, absent models and multiple currencies. Commands/results in repair records. Rollback withdraws this round’s code only; original sources/price snapshots retained. Native records permit source reselection.