Generate the weekly Android Broker on-call (OCE) WoW + 60-day trend telemetry report as a polished self-contained HTML file. Use this skill for the weekly OCE rotation when asked to "produce the OCE report", "weekly on-call report", "WoW telemetry report", "weekly broker health report", or "generate this week's on-call summary". Pulls from `android_spans` materialized views, attributes regressions/improvements to PRs in `broker/` and `common/`, and writes to `$env:USERPROFILE\android-oce-reports\oncall-wow-report-YYYY-MM-DD.html` (outside the workspace so reports are never committed).
Installation
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Generate the weekly Android Broker on-call (OCE) WoW + 60-day trend telemetry report as a polished self-contained HTML file. Use this skill for the weekly OCE rotation when asked to "produce the OCE report", "weekly on-call report", "WoW telemetry report", "weekly broker health report", or "generate this week's on-call summary". Pulls from `android_spans` materialized views, attributes regressions/improvements to PRs in `broker/` and `common/`, and writes to `$env:USERPROFILE\android-oce-reports\oncall-wow-report-YYYY-MM-DD.html` (outside the workspace so reports are never committed).
OCE Weekly Report
Produce the weekly Android Broker on-call (OCE) telemetry report as a self-contained HTML file at $env:USERPROFILE\android-oce-reports\oncall-wow-report-YYYY-MM-DD.html (where YYYY-MM-DD is the end-date of the rolling 7-day window — see "Inputs to confirm" §1). Writes to the user's home folder, outside the workspace, so reports never accidentally get committed.
The output mirrors the structure of the canonical template at assets/templates/report-template.html. The Step 1 bootstrap script copies the template into ~/android-oce-reports/oncall-wow-report-<end-date>.html, stamps the resolved rolling-7-day window into the title / meta line / Generated banner, and you edit it in place from there. Do not redesign the layout each run.
Before writing any KQL, read assets/docs/kusto-cheatsheet.md. It captures the canonical view names, helper functions, the HLL device-count gotcha, week-alignment rules, and ready-to-paste query templates — distilled from the production Android Broker Dashboard.
Canonical layout — a real prior-week report kept verbatim. Edit in place (replace dates / values / verdicts / PR links); do not restyle. See template-readme.md for what to change vs leave alone.
Per-card checklist for the deep code-attribution block (Originator / Top throw site / Wrapper / Caller hot-spots / Underlying cause / Top error_messages / Likely PRs / Next step)
Bucket all error codes into 60-day regression / spike / improvement / flat. Run with --metric=devs AND --metric=reqs. Pass --end=YYYY-MM-DD (the Sunday that OPENS the current in-progress week, exclusive — i.e. startofweek(today), printed by bootstrap as "Trend delta cutoff") to exclude the partial week from the delta math, plus --include-partial-end to still chart it as the final bar. --summary suppresses the verbose header; --json=<path> emits a structured sidecar for programmatic consumption.
Roll up 7-dim attribution slices for spike-attribution cards. Supports BOTH --union <file.json> (preferred for 2-week WoW; pairs with attr-union-by-dim.kql) AND legacy --label=<dim> file.json per-dim mode. Auto-detects the array-form schema produced by assets/scripts/run-kql.ps1 — no schema-transformer step needed.
Parallel git log -S + --grep across broker/ + common/ for a class/method symbol, with PR numbers + URLs. Run only after the Originator pre-check has identified a specific throw-site class — the unscoped 4-week PR window is small enough (<30 PRs) to scan with plain git log first.
Pre-publish validator. Catches stale tokens, devs/reqs leaks, mojibake (U+FFFD), unbalanced <div> depth in Section 2 (the nested-callout bug), KPI/trend sparkline coverage, code-attribution depth, layout-guard CSS presence, and suspicious low-peak fabricated data-trend arrays. Run as part of Step 7.
Direct-REST Kusto helper — drop-in fallback for the Azure Kusto MCP server when the MCP times out (frequent on per-error-code queries). Acquires a token via az, POSTs to /v2/rest/query, writes a JSON file the JS helpers can consume directly.
Bootstrap a new report from the canonical template. Auto-computes the rolling 7-day window (curEnd = today UTC; override with -EndDate YYYY-MM-DD), creates _data/<end-date>/, prunes _data folders older than 60 days, stamps the resolved window into the <title> / <div class="meta"> block / Generated banner, and detects "unfilled template stub" vs "real prior report" collisions using a template-only sentinel token (see § collision detection).
Optional Playwright-based layout smoke test. Renders the report at 1400 px viewport, captures a full-page screenshot under ~/android-oce-reports/_visual/, and runs DOM-based overflow + adjacent-card-gap detection. Catches the rendered-layout bugs (text bleed, cards touching) that pure HTML/CSS validation can't see.
Inputs to confirm with the user
⚠️ Do NOT ask the user for a reporting date by default. The skill uses a rolling 7-day window ending at start-of-day UTC on the invocation day so the most recent complete days are always captured. assets/scripts/bootstrap-report.ps1 computes and stamps the window automatically. The previous "confirm the Sunday bucket with the user" flow silently produced stale windows (e.g. a Thursday run emitted a 4-day partial window and dropped the last complete week) — that failure mode is what this section explicitly guards against.
Reporting window — do NOT ask. Run bootstrap-report.ps1 with no arguments; it resolves the window silently and prints:
Resolved reporting window (UTC): # example values for a run on 2026-07-15
Last 7 days: 2026-07-08 -> 2026-07-15 (exclusive upper bound)
Baseline: 2026-07-01 -> 2026-07-08
60-day trend: 2026-05-16 -> 2026-07-15 (literal 60d ending today; chart includes current partial week)
Trend delta cutoff: weeks < 2026-07-12 (startofweek(curEnd); pass as bucket-trends.js --end)
These dates are stamped into the report's <title>, <div class="meta">, and Generated banner during bootstrap — you do not hand-edit them (see assets/templates/template-readme.md § Date fields).
Override only when the user explicitly requests a non-default window. Signals: "the week of X", "as of last Friday", "the report from three weeks ago", "in-progress data", "just today's numbers". Then use:
.\bootstrap-report.ps1 -EndDate 2026-07-02 # e.g. reproduce the report as of Jul 2
-EndDate is the exclusive upper bound of the current window (curEnd); the script derives curStart = curEnd - 7d and prevStart = curEnd - 14d deterministically. -EndDate must be today or earlier — future dates are refused.
Supported override = -EndDate only (shifts the window end). The 7-day primary span, the 7-day baseline, and the 60-day trend form a fixed frame that moves rigidly with -EndDate. There is no custom start-date and no arbitrary-span flag — a request like "last 30 days" or "from Jun 1 to Jun 20" is not expressible via a single parameter. To produce a longer or custom span you must edit the KQL window placeholders (<CUR_START> / <CUR_END> / <PREV_START>; the baseline end is always <CUR_START>, so there is no separate <PREV_END> token) by hand. If the user asks for a non-7-day span, say so up front rather than silently emitting a 7-day report.
Comparison baseline — auto-computed as the immediately-prior 7 days ([curEnd - 14d, curEnd - 7d)). No user input.
60-day trend window — auto-computed as the literal last 60 days ending today ([curEnd - 60d, curEnd)), so both bounds move with -EndDate. The section is still Sun-Sat weekly-bucketed (Kusto startofweek() is Sunday-aligned) because the trend needs stable weekly denominators, but the final bar is the current in-progress (partial) week — the chart ends today. Regression/improvement delta classification is still computed on complete weeks only (bucket-trends.js --end=startofweek(curEnd) --include-partial-end); a partial week as "last" would read as a fake −99% improvement, so it is charted but excluded from the delta math. bootstrap-report.ps1 prints both the 60d data window and the "Trend delta cutoff" (= startofweek(curEnd)).
Output filename — $env:USERPROFILE\android-oce-reports\oncall-wow-report-YYYY-MM-DD.html where YYYY-MM-DD is the resolved curEnd. Example: run on 2026-07-09 (default) → oncall-wow-report-2026-07-09.html. User-scoped, outside the workspace. The filename date always matches the meta-line "Last 7 days" end-date; validate-report.ps1 check #11 asserts this.
Kusto note (60-day trend section only):startofweek() is Sunday-aligned, so startofweek('2026-05-09') == 2026-05-03T00:00:00Z. When authoring weekly-bucketed queries (60-day trend, wow-table-sparkline-series.kql), verify by printing the distinct startofweek(EventInfo_Time) values from your first query. Off-by-one-week is the #1 silent error in weekly-bucket queries.
Required sections (in order)
Top-line health KPIs — total requests, total devices, silent-auth reliability %, interactive reliability %, p95 latency on the hot spans. WoW delta on each. Inline SVG sparklines.
Things that need attention this week — callouts:
Denominator caveat — explain any large total-spans device-count shift caused by span-emission changes (e.g. goAsync() refactors). Always state which denominator the report uses (auth-only: SilentAuthStats ∪ InteractiveAuthStats).
🔴 WoW regressions (last 7 days) — one callout listing every code/type that moved sharply WoW, sorted by current-week device count descending. Built from the union of (a) the standard WoW table and (b) assets/queries/wow-movers.kql so small-but-recent spikes appear in the same list as the high-volume ones. Each row uses the .item flat-row pattern (see assets/templates/template-readme.md § "Section 2 callouts"): name + inline metric chips + tags pushed right + one-line body + optional foot with Attribution card → link. Section 2 rows are at-a-glance only — do not duplicate the dim slicing / PR analysis / detailed verdict here; that belongs in the Section 4 spike-attribution card. Each row carries tags: NEW (first appeared this week or last), 60d↑ (also rising on 60d), and an originator chip (broker / eSTS / Android / env). Reader's eye prioritizes naturally by row order and tag combination — broker-tagged rows at the top demand the most attention.
Slow-burn 60-day regressions — codes/types climbing on the 60d window that are flat WoW. Anything that also moved WoW belongs in the red callout above (with 60d↑), not here. Link to the 60-Day Trend section.
Real wins this week, with PR links.
Traffic shape — flat / surge / collapse summary.
📈 60-Day Trend Analysis — built from the ErrorStatsMetrics materialized view over the literal last 60 days ending today (final bar = current in-progress week). Run the bucketing pipeline FOUR times — the cross-product of {error_code, error_type} × {devices, requests} — and union the regression sets. An entry (code OR type) is flagged if it regresses on either metric. Deltas are computed on complete weeks only; the partial current week is charted but excluded from classification.
% of devices affected (devicesHit / authActiveDevices) — catches errors hitting more users.
% of requests affected (errRequests / authTotalRequests) — catches per-device retry storms (fewer users, more traffic per user). The previous report would have missed kdfv2_key_derivation_error (262 → 5,374 requests on ~57 devices) without this dim.
Categories: True 60d regression / Ephemeral 60d spike (peak-then-recover) / True 60d improvement / Flat. Every rising entry — whether error_code or error_type — gets the same Spike Attribution + Code Attribution treatment (Step 4 / Step 5).
Always apply MergeUiRequiredExceptions(error_type) before bucketing on type; otherwise the 6+ string variants of UiRequiredException will each be tracked separately and skew the buckets.
🔎 Spike Attribution — one card per WoW regression AND per 60-day regression, for both error_code and error_type regressions. Each card slices on all 7 dimensions (broker version, span, active broker pkg, calling app, account type AAD/MSA, shared-device mode, client SKU). Each card ends with a deep Code Attribution block (see Step 4 for the required fields) and a Traffic Attribution verdict.
🚚 Traffic Attribution — top-level section listing every error whose spike is fully or partly explained by traffic volume from a specific calling app, rather than a code regression. If none qualify this week, render the section with an explicit "None this week" note.
Error codes — WoW with stable denominator — full table with Δ requests % and Δ devices % columns and the 60d sparkline.
Error types — WoW with stable denominator — full table, same columns and rigor as the error-codes table (Δ requests %, Δ devices %, 60d sparkline, status pill). Any regressing type also gets a spike-attribution card in Section 4. For composite types (e.g. ClientException is the umbrella for many sub-codes), include a decomposition card that breaks the WoW Δ down into the top 3 contributing sub-codes — so a ClientException −5 pp drop is explicitly attributed to e.g. −8.5 pp timed_out_execution + −3.4 pp unknown_authority + −0.15 pp illegal_argument_exception.
📊 Traffic analysis — total requests/devices (WoW + 60d), top calling apps, top spans, requests-per-device ratio per error and overall (a rising ratio = retry storm; a falling ratio = caching gain), sampling-rate change indicator.
Latency — p50/p95/p99 by hot span.
Broker version adoption — week-over-week version share.
Appendix — query list and methodology.
Step-by-step workflow
Step 1 — Bootstrap the new report file from the template
This skill ships with a canonical template at assets/templates/report-template.html (a real prior report kept as the reference layout). Use assets/scripts/bootstrap-report.ps1 to handle all the boilerplate (rolling-window computation, _data/<end-date>/ directory, header stamping, retention-pruning, collision detection):
Resolves the rolling 7-day window from the system clock in UTC (curEnd = today, curStart = curEnd - 7d, prevStart = curEnd - 14d) — or from -EndDate if passed.
Creates ~/android-oce-reports/oncall-wow-report-<curEnd>.html from the canonical template.
Creates ~/android-oce-reports/_data/<curEnd>/ for raw KQL JSON payloads.
Stamps the resolved window into the report's <title>, <div class="meta"> block, and Generated banner — you never hand-edit header dates. The resolved window is echoed in the report header for transparency.
Prunes _data/<old-end-date>/ folders older than 60 days so the cache doesn't accumulate.
Collision detection (fail-safe): an existing same-day report is silently re-bootstrapped only when it is positively identified as an unpopulated stub — it still carries the OCE-UNPOPULATED-STUB sentinel that bootstrap injects and its first KPI still equals the template's value. Anything else (sentinel removed, or KPIs edited) is treated as real work: HARD HALT, exit 2, requiring -Force to overwrite. validate-report.ps1 refuses to pass a report that still carries the sentinel, so a published report can never be misclassified as a stub.
Edit the bootstrapped file in place — the template ships as a real prior-week report (not a tokenized skeleton). Walk top-to-bottom and replace every prior-week date / KPI value / table row / verdict / PR citation with current-week data. The CSS, sparkline JS, section ordering, and attribution-card markup are canonical — do not redesign them. See assets/templates/template-readme.md for the full guide on what to change vs leave alone, the sparkline color palette, the CSS class reference, and the two v8 layout traps.
⚠️ UTF-8 trap — DO NOT use PowerShell @'...'@ heredocs to compose HTML content containing emojis, em-dashes, arrows, or middle dots. PowerShell silently strips multi-byte UTF-8 characters when piping heredocs to Set-Content / Out-File. Use Node.js (fs.writeFileSync), [IO.File]::WriteAllText($path, $text, [System.Text.UTF8Encoding]::new($false)), or explicit Unicode-pair literals ([char]0xD83D + [char]0xDCCA for 📊) instead. This trap cost ~30 min in v8 and required a full emoji-restoration pass — every callout icon, every section header emoji, every arrow link had to be re-injected. The validator's U+FFFD check catches the worst case (mojibake replacement char) but cannot detect characters that were silently stripped to nothing.
Mark any unfinished card or table cell with the literal sentinel EXAMPLE CONTENT BELOW inside an HTML comment — the final-pass validator (Step 7) greps for it.
If the template ever needs structural improvements (new section, new card style, etc.), update assets/templates/report-template.html in the skill folder and commit it so future weeks inherit the change.
Step 2 — Pull WoW reliability data
Use the Kusto MCP tool against:
Cluster:https://idsharedeus2.kusto.windows.net
Database:ad-accounts-android-otel
Always prefer the canonical materialized_view('XxxMetrics' or 'XxxUpdated') variants — these are what the production dashboard uses, are pre-aggregated and HLL-bucketed, and avoid the 240 s MCP timeout that plain android_spans queries hit. Full schema, gotchas, and query templates: assets/docs/kusto-cheatsheet.md.
Fallback when the Kusto MCP times out: use assets/scripts/run-kql.ps1. It acquires a token via az account get-access-token, POSTs directly to /v2/rest/query, and writes the result as a JSON file the JS helpers (bucket-trends.js, summarize-attribution.js) can consume directly. The skill's MCP-vs-REST switch is roughly: try the MCP once; if it returns McpError -32001 (timeout), switch to the REST helper for the rest of the run. Run multiple queries in parallel via PowerShell Start-Job:
Time filter: always use EventInfo_Time on materialized views. Use PipelineInfo_IngestionTime only on raw android_spans.
Three rules that will silently corrupt your data if violated (full detail in the cheatsheet):
Distinct devices are HLL-encoded. Use dcount_hll(hll_merge(countDevicesHll)), never sum(countDevices). Summing double-counts every device that appears in more than one row.
Apply the dashboard helper functions so this report agrees with the dashboard: MergeAccountType(account_type), MergeIsSharedDevice(is_shared_device), MergeUiRequiredExceptions(error_type).
Auth-only denominator for reliability %s: sum countRequests from SilentAuthStatsAllRequestsMetrics ∪ InteractiveAuthStatsAllRequestsMetrics — not total broker spans. Total span counts are sensitive to goAsync() / receiver refactors and will give false WoW reliability swings.
Step 3 — Pull 60-day trend
Don't pre-filter to a hand-picked top-N list — small-but-rising errors (e.g. null_pointer_error at ~67K devices) will fall off and never show up in the trend section. Instead pull every error code and every error type with a meaningful baseline across the window, then bucket each.
3a. Per-error-code trend
Use assets/queries/60d-trend-codes.kql (template; replace <TREND_START> and <TREND_END> tokens. <TREND_START> = curEnd − 60d and <TREND_END> = curEnd (today), exclusive — the literal last 60 days. bootstrap-report.ps1 prints the resolved values):
materialized_view('ErrorStatsMetrics')
| where EventInfo_Time >= datetime(<TREND_START>) and EventInfo_Time < datetime(<TREND_END>)
| where isnotempty(error_code) and error_code != 'success'
| summarize errs = sum(countOverall),
devs = dcount_hll(hll_merge(countDevicesHll))
by week = startofweek(EventInfo_Time), error_code
| order by error_code asc, week asc
Do NOT filter the partial in-progress week here. The chart wants it as the final bar (the window ends today). The partial week is excluded from the regression/improvement delta math by bucket-trends.js via --end=<TREND_CLASS_END> --include-partial-end (see 3c), not at the source — a partial week driving the delta would read as a fake −99% improvement, which is exactly why classification and display are split in the JS.
3b. Per-error-type trend (same rigor)
materialized_view('ErrorStatsMetrics')
| extend unified_error_type = MergeUiRequiredExceptions(error_type)
| where EventInfo_Time >= datetime(<TREND_START>) and EventInfo_Time < datetime(<TREND_END>)
| where isnotempty(unified_error_type)
| summarize errs = sum(countOverall),
devs = dcount_hll(hll_merge(countDevicesHll))
by week = startofweek(EventInfo_Time), unified_error_type
| order by unified_error_type asc, week asc
MergeUiRequiredExceptions is mandatory — without it the 6+ string variants of UiRequiredException (raw, fully-qualified, com.microsoft.identity.common.exception.*) each show as separate rows and skew the buckets.
3c. Run the bucketer 4 times (cross-product of {code, type} × {devices, requests})
bucket-trends.js defaults to grouping by error_code. For the type runs you MUST pass --key=unified_error_type so it picks up the right column from the type-trend JSON.
# Error codes — by devices, then by requests.
# TREND_START = curEnd - 60d (literal 60d start)
# TREND_CLASS_END = startofweek(today) ("Trend delta cutoff" printed by bootstrap)
# --include-partial-end charts the current partial week while excluding it from deltas.
node .github\skills\oncall-weekly-telemetry-report\assets\scripts\bucket-trends.js <codes.json> --start=<TREND_START> --end=<TREND_CLASS_END> --include-partial-end
node .github\skills\oncall-weekly-telemetry-report\assets\scripts\bucket-trends.js <codes.json> --start=<TREND_START> --end=<TREND_CLASS_END> --include-partial-end --metric=reqs
# Error types — by devices, then by requests (note --key)
node .github\skills\oncall-weekly-telemetry-report\assets\scripts\bucket-trends.js <types.json> --start=<TREND_START> --end=<TREND_CLASS_END> --include-partial-end --key=unified_error_type
node .github\skills\oncall-weekly-telemetry-report\assets\scripts\bucket-trends.js <types.json> --start=<TREND_START> --end=<TREND_CLASS_END> --include-partial-end --key=unified_error_type --metric=reqs
--end is <TREND_CLASS_END> = startofweek(today) (exclusive) — the Sunday that opens the current in-progress week. Weeks at or after it (the partial current week) are excluded from delta classification; --include-partial-end keeps that week in the emitted series so the chart ends today. The script also auto-detects partial end-buckets and warns if --end is omitted, but passing it explicitly is safer.
Take the union of all four regression sets. Both error_code and error_type regressions get a spike-attribution card in Step 5.
It will print regression / spike / improvement / flat buckets, sorted by peak. The thresholds (in case you need to tune):
True 60d regression:delta > +15% and trajectory is monotonic-ish (no single-week spike dominating).
Ephemeral 60d spike: peak week is ≥3× the mean of the surrounding weeks (peak-then-recover shape).
True 60d improvement:delta < −15%.
Flat: otherwise.
Codes/types with peak weekly devices < 10K (or peak weekly requests < 100K when --metric=reqs) are filtered out (--peak-floor=N to override).
Why both axes matter:
codes × requests: in v5, kdfv2_key_derivation_error spiked +1,951% on requests across only ~57 devices — a per-device retry storm device-only bucketing would have missed.
types × either:error_type is the umbrella (e.g. ClientException, ServiceException, UiRequiredException) — a moving type that doesn't map cleanly to one moving code is a strong signal of a new sub-code being introduced or an existing one being reclassified (the v5 ClientException −10% drop was driven by timed_out_execution reclassification under PR #141, which would have been invisible from the codes table alone).
Always present side-by-side WoW tables for BOTH error_code AND error_type with Δ requests % and Δ devices % columns; flag any row where either crosses threshold.
The 60d bucketer's --peak-floor=10000 exists for good reason (otherwise the 60d regression list would be 200+ tiny noise codes), but it silently drops every code whose absolute weekly volume stays under 10K — even if that code is brand-new or just spiked 5× WoW. Real examples this skill has missed in the past:
Failed to parse JWT — went 7 → 32 → 54 → 46 → 55 → 892 → 3,461 over 7 weeks (2-week-old NEW spike, real broker code in IDToken.parseJWT:38). Never crossed the 10K floor.
Code:-11 — sat at ~1,030 devs/week for 7 weeks then jumped to 2,433 (+165% WoW). Sub-floor.
SSLHandshakeException — devices flat at 260 but requests +186% WoW (per-device retry storm). The bucketer's reqs-axis floor (100K) just barely captures it but the device floor doesn't.
// inputs: <CUR_END> = curEnd (exclusive), <CUR_START> = curEnd - 7d,
// <PREV_START> = curEnd - 14d. Printed by bootstrap-report.ps1.
// floor: cDev>=500 OR cReq>=5000 move: |Δd|>=25% OR |Δr|>=50% OR new-this-window
Run it twice — once for error_code, once for error_type. Merge its output rows into the same 🔴 WoW regressions callout as the standard WoW table (sorted by current-week device count descending). Tag rows that came in via this pass with NEW if they were absent or near-zero in the prior week. Do not render this as a separate "emerging" callout — the size split is implementation detail; readers prioritize naturally by absolute device count + originator chip.
For each WoW mover (regardless of size), you still owe the full Code Attribution treatment (Step 4). The dim-slicing pass (Step 5) is allowed to be deferred for sub-1K-device spikes if the throw-site + dominant message already pin the originator unambiguously — but say so explicitly in the card ("dims not yet sliced — file the bug first; pull dims if it persists").
Step 4 — Code attribution (deep PR correlation)
⚠️ HARD RULE — Originator pre-check. Before claiming Originator: Broker on any card, you MUST run assets/queries/error-message-and-location.kql for that error code (or type) and read (a) the throw-site stack and (b) the top 3 error_message strings. Most broker error codes flow through common/ExceptionAdapter.{getExceptionFromTokenErrorResponse, exceptionFromAuthorizationResult, clientExceptionFromException} — which intentionally bridge eSTS responses into broker exceptions. If the throw site is in any of those three methods AND the error_message starts with AADSTS, the originator is eSTS, not broker. See the AADSTS reference table in assets/docs/kusto-cheatsheet.md. Cards that skip this step must be marked low-confidence, not high.
Window: use the FULL 7-day rolling window (<CUR_START> → <CUR_END>) on PipelineInfo_IngestionTime, NOT a narrower 3–5 day slice — low-volume types (e.g. SSLHandshakeException, IntuneAppProtectionPolicyRequiredException) routinely return zero rows in a sub-window slice. If a code/type still returns nothing, fall back to the prior 14 days (<PREV_START> → <CUR_END>) before declaring "no data".
For every regression card, the Code Attribution block must populate the following fields. Shallow PR-citation only is not acceptable. Use assets/docs/code-attribution-template.md as the per-card checklist.
Field
What goes in it
How to find it
Originator
Where the error physically originates: broker code / common / Android system (WebView / Conscrypt / Keystore) / 3rd-party lib (Nimbus JWT, okhttp) / eSTS server / environmental (enterprise TLS interception). Use the colour-coded origin-tag spans (origin-broker, origin-android, origin-thirdparty, origin-env).
Grep the error string across broker/, common/, msal/. If no match, it's not our code — search the Android SDK or call out as eSTS-returned.
Top throw site
Fully-qualified file:line where the exception is constructed, plus the % of cases that throw from this single site.
Pull error_location / stack-prefix from android_spans for the spiking error code (one targeted query, narrow time window). Cite the dominant site.
Wrapper
Broker/common code that catches the originator's exception and re-throws it as the user-visible error code. Often IDToken.parseJWT(), ServiceException(...), ExceptionAdapter.exceptionFromAuthorizationResult().
Walk up the stack from the throw site — check for try { ... } catch (X e) { throw new Y(...); } patterns in broker/common.
Caller hot-spots
Top 1–3 callers of the wrapper, with device counts. Helps identify the specific code path the regression flows through.
android_spans slice by error_location (or error.stack_trace first frame inside our code).
Underlying cause
The proximate cause one level deeper (e.g. "99% CertificateException from TrustManagerImpl.verifyChain", "84% no_such_algorithm from ProviderFactory.getMessageDigest").
android_spans slice by error.cause or error_message first 80 chars.
Top error_messages
Top 3–5 distinct error_message strings with counts. Often reveals the 3rd-party library or environmental signal (e.g. net::ERR_SSL_PROTOCOL_ERROR, Zscaler-issued cert names).
summarize count() by tostring(error_message) on raw android_spans filtered to the spike.
Likely PRs
1–3 PRs with confidence rating (high / medium / low / none), full GitHub URL, commit SHA, author, AB#, and a 1-sentence why-it's-the-suspect justification (not just the title). Use the pr-card markup.
See PR-grep below. Cite confidence honestly — "none" is a valid verdict for environmental errors.
Next step
Concrete action with a named owner: who runs the next slice, who files the bug, what flight to flip, what correlation IDs to pull.
Pulled from PR authors / CODEOWNERS for the affected file.
PR-grep workflow
Read the full PR window first, then reason — don't --grep blind. The 4-week window across broker/ and common/ typically returns <30 PRs total, small enough to read end-to-end. Targeted --grep matches will miss PRs whose titles don't mention the error string (most of them). The recommended order is:
Run plain git log on both repos for the 4-week window. Read the resulting list end-to-end before any greps.
Cross-reference titles + dates against the Originator pre-check throw-site class.
Only when you have a specific symbol to chase (e.g. the throw-site class identified in step 2), reach for find-suspect-prs.ps1 to do the symbol-targeted parallel pickaxe + grep.
The historical mistake (pre-v8) was to jump straight to find-suspect-prs.ps1 without reading the window first, which silently dropped PRs whose titles didn't mention the symbol.
# Step 1: read the full 4-week window
cd c:\Users\shjameel\Repos\android-complete\broker
git --no-pager log --since='<windowStart>' --until='<windowEnd>' --pretty=format:'%h | %ai | %an | %s' --no-merges
cd ..\common
git --no-pager log --since='<windowStart>' --until='<windowEnd>' --pretty=format:'%h | %ai | %an | %s' --no-merges
For each candidate PR, read the diff to confirm it touches the throw site / wrapper class identified in the Originator pre-check. Don't cite a PR just because the title mentions a related concept.
# Step 3 (optional): symbol-targeted focused follow-up. Use ONLY after step 1 gave
# you a specific class/method name to chase from the Originator pre-check.
# Searches both repos in parallel via `git log -S` (pickaxe on diff) AND `--grep` (subject).
# Returns a unified table: repo | date | author | sha | PR# | URL | subject.
.\.github\skills\oncall-weekly-telemetry-report\assets\scripts\find-suspect-prs.ps1 `
-Symbol 'ExceptionAdapter' -Since 2026-04-01 -Until 2026-05-09
For errors with no broker code in the stack (Android system errors like Code:-10/Code:-11, OEM-specific keystore failures, eSTS-returned codes, environmental TLS interception), explicitly cite "⚪ None — not in scope" with confidence none, and explain why in the why-it's-the-suspect line. Do not invent broker PRs to fill the slot. Tag these errors as environmental or non-broker so they're tracked but don't page.
Step 5 — Spike attribution dimensions
Coverage rule: every error_code AND every error_type that lands in either the WoW regression list OR the 60-day regression list MUST get a spike-attribution card. No silent skips.
ErrorStatsMetrics already carries account_type and is_shared_device (use the MergeAccountType / MergeIsSharedDevice helpers to normalize) — so you do not need a fallback to raw android_spans for these dims. Earlier versions of this skill claimed otherwise; that was wrong. The only dim that requires android_spans is DeviceInfo_OsVersion (OEM/version slicing).
Slice on all 7 dimensions for each spike. Preferred for 2-week WoW attribution: one union query that covers all 7 dims for all regressions in a single round-trip — see assets/queries/attr-union-by-dim.kql. Typical payload for 8 codes × 2 weeks × 7 dims is ~800 KB, well under the MCP limit. Pipe the result into summarize-attribution.js --union <file.json> (which prints per-dim top-N share + Δ devices + Δ requests for every code). Fall back to the per-dim form (attr-codes-by-dim.kql) only when (a) you need a wider time window, or (b) the union response exceeds payload size.
For error_type cards, swap error_code in (codes) for unified_error_type in (types) and aggregate by the MergeUiRequiredExceptions(error_type) extension — otherwise everything else is identical.
Low-volume fallback (extends Step 4's pre-check fallback to the 7-dim union): when a code/type returns sparse dim rows in the 7-day rolling window — typical for sub-1k-device entries like TimeoutCancellationException, JsonSyntaxException, kdfv2_key_derivation_error — widen the union query to 14 days (use <PREV_START> as the lower bound so the window becomes [curEnd − 14d, curEnd)) before declaring "broad — needs targeted slice". The added week of context usually surfaces enough rows to compute concentration percentages. If a code STILL has no concentration after 14 days, mark every dim cell as "not sliced — sub-window volume; file the bug first, slice on persistence" — do NOT fabricate "Broad" verdicts.
#
Dimension
Source
Cross-check
1
Broker version
ErrorStatsMetrics group by broker_version
Cross-reference BrokerAdoptionStatsUpdated to see if the version's request share also moved that week — if yes, the spike is rollout-driven, not code-driven
2
Span name
ErrorStatsMetrics group by span_name
A single span hosting >60% of the error → strong code-path signal
3
Active broker package
ErrorStatsMetrics group by active_broker_package_name
E.g. CompanyPortal vs Authenticator vs LTW
4
Calling package
ErrorStatsMetrics group by calling_package_name
If 1–2 callers dominate, this is likely a traffic-attribution case (see Step 6)
5
Account type (AAD vs MSA)
ErrorStatsMetrics, extend t = MergeAccountType(account_type) group by t
If the split deviates significantly from fleet (~85% AAD / 15% MSA), call it out
6
Shared device mode
ErrorStatsMetrics, extend s = MergeIsSharedDevice(is_shared_device) group by s
Shared-device fleets have very different error profiles
On-demand only — slice OS-version when EITHER (a) the wrapper class is in ExceptionAdapter.clientExceptionFromException (catch-all wrapping a system exception, where the OEM/version often is the cause), OR (b) the error code is one of Code:-6, Code:-10, Code:-11, unknown_crypto_error, io_error, null_pointer_error. Otherwise mark the dim row as "not sliced this week — no OEM concentration suspected" and move on. Slicing OS-version on every card wastes a raw-spans query without changing the verdict.
Type cards have one extra required dimension: sub-code decomposition
Because error_type is an umbrella over many error_code values, every error_type regression card MUST also include an 8th dimension: sub-code breakdown showing the top 3–5 error_codes rolled up under that type, with their device counts and Δ vs prior week. This lets the reader see whether the type-level move is driven by one sub-code or many — and routes the deep Code Attribution work to the right sub-code.
let curEnd = datetime(<CUR_END>);
let curStart = datetime(<CUR_START>);
let prevStart = datetime(<PREV_START>);
let target_types = dynamic(['ClientException', 'ServiceException']);
materialized_view('ErrorStatsMetrics')
| extend unified_error_type = MergeUiRequiredExceptions(error_type)
| where EventInfo_Time >= prevStart and EventInfo_Time < curEnd
| where unified_error_type in (target_types)
| extend week = iff(EventInfo_Time >= curStart, curStart, prevStart)
| summarize devs = dcount_hll(hll_merge(countDevicesHll)),
errs = sum(countOverall)
by week, unified_error_type, error_code
| order by unified_error_type asc, week asc, devs desc
Cite the dominant sub-codes inline in the type card's verdict (e.g. "ClientException −10.2% drop is dominated by −8.5 pp timed_out_execution + −3.4 pp unknown_authority") and link to those sub-codes' own attribution cards. The deep Code Attribution block (Step 4) for the type card itself focuses on the wrapper / catch-and-rethrow path that defines the type (e.g. BaseException.java, ServiceException.java constructors), not on each sub-code.
Feed the union JSON output into the summarizer (one round-trip):
# Union mode (preferred). attr-union.json comes from attr-union-by-dim.kql.
node .github\skills\oncall-weekly-telemetry-report\assets\scripts\summarize-attribution.js `
--union attr-union.json --top=5
# For type cards, add --key=unified_error_type
Legacy per-dim mode (one JSON per dimension) is still supported for the rare wider-time-window case:
80% in a single value → strong attribution (one root cause)
60–80% → medium attribution
< 60% → broad / cross-cutting → say so explicitly, don't fabricate a single cause
Step 6 — Traffic analysis + traffic attribution
Do this section in three parts. Traffic changes (up or down) need the same level of root-cause reasoning as error spikes — a uniform "−9% requests across all top apps with flat devices" is not a satisfactory verdict on its own; explain why.
6a. Top-line traffic shape. Compare WoW and 60d for both totals and per-segment:
materialized_view('BrokerAdoptionStatsUpdated')
| where EventInfo_Time > ago(70d)
| summarize totalReq = sum(countRequests),
totalDev = dcount_hll(hll_merge(countDevicesHll))
by week = startofweek(EventInfo_Time)
| order by week asc
For each of the following, report direction + magnitude:
Total requests (WoW %, 60d %)
Total devices (WoW %, 60d %)
Requests-per-device ratio (a drop often means a benign caching improvement; a spike often means a retry storm)
Top 10 calling apps (AppStatsUpdated) — which apps drove the change?
Top spans by request volume — did one span explode or collapse?
Sampling-rate change indicator: if total spans moved >20% but auth-only device count moved <5%, suspect a sampling/instrumentation change.
6b. Reasoning for material traffic shifts (>10% on any segment). For every span/app/active-broker that moved meaningfully WoW or 60d, run this slicing-and-correlation pass:
#
Question
How to check
1
Is the move concentrated in one span?
Slice top-10 spans by Δreq absolute and Δreq %. A >50% move on a single span almost always points to a code change (span added / removed / sampled / goAsync()-ed).
2
Is the move concentrated in one calling app?
Slice AppStatsUpdated WoW. A single app moving >20% in requests with flat devices = client-side caching/retry change in that app — escalate to that app's owners, not broker.
3
Is the move concentrated in one active broker pkg?
Slice BrokerAdoptionStatsUpdated by active_broker_package_name. AppManager (LTW) vs Authenticator vs Intune CP often diverge during a rollout.
4
Is the move concentrated in one broker version?
Cross-check against rollout share. If a span dropped −80% on 16.0.1 but is flat on 15.1.0, the cause is in the 16.0.1 diff.
5
Did anything else co-move?
A span dropping while OnUpgradeReceiver-style downstream spans also drop (SecretKeyWrapping, WrappedKeyAlgorithmIdentifier in v5) confirms a single upstream change.
For every meaningful shift, search for a causal PR in the repos likely to affect telemetry shape:
Causal PR categories that meaningfully shift traffic counts (flag any of these):
Span removed / renamed / scope-narrowed → drops the span's count to zero or partial
goAsync() / BroadcastReceiver refactor → broadcast may complete before async work flushes the span (this is the v5 PR #88 / OnUpgradeReceiver story — call it out as a precedent)
Sampling-rate change in broker Otel* / Telemetry* exporter config or common/ instrumentation → uniformly scales counts up or down across many spans
New span added in a hot path → request counts for that span jump from ~0 to material
Caller-side SDK change (MSAL/MSAL_CPP/OneAuth release) that batches or caches requests → uniform per-app request drop with flat devices
Flight rollout (ECS) that gates a code path on/off → bursty changes in a specific span on specific dates
Cite the suspect PR(s) with the same confidence ratings used in Code Attribution (high / medium / low / none) and the same pr-card markup. If you can't pin one down, say so explicitly — "uniform 5–22% per-app request drop with flat devices, no telemetry-platform PR identified, suspect caller-side SDK change in MSAL release X.Y" is acceptable; "traffic is flat" without checking is not.
6c. Per-error traffic attribution (is the error spike traffic-driven?). For every error code flagged in Step 5 as a regression, additionally check whether the spike is traffic-driven rather than failure-rate-driven:
let curEnd = datetime(<CUR_END>);
let curStart = datetime(<CUR_START>);
let prevStart = datetime(<PREV_START>);
let target_code = "<error_code>";
materialized_view('ErrorStatsMetrics')
| where EventInfo_Time >= prevStart and EventInfo_Time < curEnd and error_code == target_code
| extend week = iff(EventInfo_Time >= curStart, curStart, prevStart)
| summarize errs = sum(countOverall),
devs = dcount_hll(hll_merge(countDevicesHll))
by week, calling_package_name
| order by week asc, devs desc
If the spike is concentrated in a single calling app whose overall request volume also rose that week (cross-check AppStatsUpdated), and the per-request failure rate is essentially flat, classify the spike as a traffic-attribution case rather than a code regression:
Example: "no_account_found +60% devices this week is fully explained by Outlook's request volume rising 65% — the per-Outlook-request failure rate is unchanged. No broker code change is implicated."
Add a top-level 🚚 Traffic Attribution section that lists every error matched to a traffic-driven origin, mirroring the Code Attribution section. Each card must include: the dominant calling app(s) with their WoW request-volume delta, the per-app per-request failure rate (now vs prior — show it's flat), and the recommended owner to route to (typically the calling app's team, not broker). If no errors qualify in a given week, render the section with an explicit "None this week" note rather than omitting it.
Step 7 — Validate & write
Run the bundled validator FIRST — it covers all the silent-failure cases this skill has tripped on in the past:
.\.github\skills\oncall-weekly-telemetry-report\assets\scripts\validate-report.ps1
# defaults to most-recent oncall-wow-report-*.html under ~/android-oce-reports/
# pass -Path explicitly to validate a specific file
The validator hard-fails on:
Stale {{...}} tokens or EXAMPLE CONTENT BELOW / EXAMPLE_* sentinels.
devs / reqs in user-facing text (KQL inside <pre><code> is exempted).
U+FFFD replacement characters (catches mojibake from emoji edits).
Unbalanced <div> depth in the Section 2 attention block (catches the inception-style nested-callout bug from past runs).
A second callout opening before the previous one closes (nested-callout sanity check).
Chartless KPI grid — if more than half the .kpi tiles lack a data-spark element (catches the v7 regression where the body was rebuilt without sparklines). Also warns when total chart count (sparks + trends + inline svgs) is < 15.
Code-attribution depth — each .attr-card's "Code attribution" block must contain an Originator row (proxy for the full 8-field structure: Originator / Top throw site / Wrapper / Caller hot-spots / Underlying cause / Top error_messages / Likely PRs / Next step). Catches the v7-third-pass regression where cards shipped with a pr-list-only stub.
Attribution-card layout guards (v8) — the CSS must define .attr-card { margin-bottom: 16px } AND .dim-row overflow rules (text-overflow: ellipsis + min-width: 0). Catches the "cards touching" and "text bleeding out of dim boxes" regressions from a stale <head> block.
Fabricated-sparkline heuristic (v8) — warns when a data-trend array's peak value is < 100 (almost certainly hand-rolled rather than sourced from real data). See assets/queries/wow-table-sparkline-series.kql for the canonical KQL that pulls real 8-week series for every code in the WoW tables. Its <SPARK_START> / <SPARK_END> tokens are the last 8 complete Sun-Sat weeks (<SPARK_END> = startofweek(today), exclusive) — deliberately distinct from the trend-chart's <TREND_START> / <TREND_END> (literal last 60 days ending today). Per-row sparklines stay on complete weeks so a partial final point doesn't create a misleading dip in every WoW row.
Then:
Run the visual smoke test (recommended) — catches rendered-layout bugs that pure HTML/CSS validation can't see:
.\.github\skills\oncall-weekly-telemetry-report\assets\scripts\visual-smoke.ps1
# Opens the report at 1400px in headless Chromium via Playwright, captures a
# full-page screenshot to ~/android-oce-reports/_visual/, and runs DOM-based
# checks for:
# - element overflow inside .dim / .attr-card (catches "text bleeding out")
# - adjacent .attr-card pairs with gap < 8px (catches "cards touching")
# First run auto-installs Playwright + Chromium into %LOCALAPPDATA%\oce-skill-playwright
Run get_errors on the HTML file (no errors expected — pure HTML/CSS).
Verify no stale phrases from prior weeks remain (Select-String for retracted hypotheses, prior week's PR numbers).
Verify every PR link in the new file is reachable (the file paths just before the link should match what git log returned).
Hard rules
Never sum(countDevices). Always dcount_hll(hll_merge(countDevicesHll)). Summing the per-row distinct count double-counts.
Always wrap view names in materialized_view('Xxx') and use the canonical Metrics/Updated variants (see cheatsheet § 2).
Never sum percentiles. Latency is a TDigest sketch — percentile_tdigest(tdigest_merge(responseTimeTDigest), N, typeof(long)) only.
Always apply MergeAccountType / MergeIsSharedDevice / MergeUiRequiredExceptions so this report agrees with the dashboard.
Confirm the week bucket label matches the user's intent before writing the rest of the queries (Sunday-aligned).
Do NOT filter the partial in-progress week at the source in the 60-day trend queries — the chart ends today and wants that partial week as its final bar. Exclude it from the regression/improvement delta math instead by running bucket-trends.js --end=<startofweek(today)> --include-partial-end: the --end cutoff drops the partial week from first/last/delta classification while --include-partial-end keeps it in the emitted series. Skipping --end (or the cutoff) would make bucket-trends.js show every error as a fake −99% improvement. The per-row wow-table-sparkline-series.kql is the exception — it keeps 8 complete weeks (<SPARK_END> = startofweek(today), with the partial week filtered at the source) so no WoW row ends on a misleading partial dip.
Never carry a numeric telemetry value forward between runs. Every KPI, table cell, delta %, device/request count, sparkline point, and verdict number must be re-pulled from Kusto for this run — never copied from a previous report, from a checkpoint/summary, from notes, or from memory. Telemetry shifts between runs and stale numbers read as fabricated. Near-miss precedent: a no_tokens_found count was about to be carried as ~23.7M when the actual current-window value was ~4.86M — a ~5× error that only the re-pull caught. If a number isn't backed by a query result file in this run's _data/<end-date>/, it does not go in the report.
Never hardcode the "Generated" date. It is the run date in UTC, auto-stamped by bootstrap-report.ps1 (which uses (Get-Date).ToUniversalTime()). If you rebuild the body programmatically, derive it live with a UTC-date formatter (new Date().toISOString().slice(0,10) in Node, [datetime]::UtcNow.ToString('yyyy-MM-dd') in PowerShell) — never paste a literal, and stay on UTC so the assembler can never stamp a different day than bootstrap-report.ps1. The v8 "Generated 2026-06-15 on a 2026-06-18 file" bug came from a hardcoded string in the assembler. (Reporting-week / baseline / 60d window dates are author-set and verified against the user's intended Sunday bucket — see template-readme "Date fields".)
Originator pre-check is mandatory. A card cannot claim Originator: Broker without first running assets/queries/error-message-and-location.kql and reading the throw site + top 3 error_message strings. If the throw site is in common/ExceptionAdapter.{getExceptionFromTokenErrorResponse, exceptionFromAuthorizationResult} AND the message starts with AADSTS, the originator is eSTS, not broker — see the AADSTS reference in assets/docs/kusto-cheatsheet.md.
WoW-movers pass is mandatory. The 60d bucketer's --peak-floor silently drops sub-10K-device codes, so assets/queries/wow-movers.kql MUST be run as a separate pass for both error_code and error_type (per Step 3d). Its output is merged into the single 🔴 WoW regressions callout, sorted by current-week device count descending, with rows tagged NEW / 60d↑ / originator chip. Do not render a separate "emerging" callout. Skipping the pass is how the Apr 26 Failed to parse JWT spike (7 → 3,461 devs over 7 weeks) hid for two reports running.
Section 2 callouts are at-a-glance, Section 4 is the deep dive. WoW / Slow-burn / Wins items in Section 2 use the .item flat-row pattern (no nested cards, no per-item left bars — the parent .callout border is the only severity affordance). Each row is a single line of metric chips + a one-line body + an Attribution card → link to the corresponding .attr-card in Section 4. Do NOT duplicate the dim slicing, PR analysis, or detailed verdict between the two sections — Section 4 is where that lives. See assets/templates/template-readme.md for the CSS class reference and the example .item markup.
Never use bash/PowerShell regex to bulk-edit balanced HTML. This skill has burned twice on regex strip scripts that ate matched-pair </div> closes, producing inception-style nested-callout bugs that take a depth-tracking script to find. If you need a structural change to the HTML, make a targeted, single-occurrence string replacement (with explicit before/after context) or rewrite the affected block end-to-end. Never run a -replace across the whole file expecting it to leave balance intact.
Denominator caveat must cite evidence, not hand-wave. If you flag a large all-spans device-count shift, run assets/queries/broker-version-share-wow.kql (single WoW snapshot) or assets/queries/broker-version-share.kql (time-series) and name the version cohort the shift moved with. Do not write "recurring telemetry-shape artifact" without backing data; if you don't have it, drop the callout.
"Recovery" still merits a PR citation. When an error pins to a single old broker version and recovers as that version retires, look for the fix PR in the version that replaced it before calling it a "natural rolloff." Often the fix is real and just under-credited.
Never report WoW-only verdicts for errors that are flat-or-down WoW but rising on 60d — always cross-check both windows.
Never page based on a regression that turns out to be a downstream of a denominator shift; always include the auth-only-denominator number alongside the all-spans number.
Always cite PRs with full GitHub URLs (the repo URL patterns above), not bare commit SHAs.
Filename collision rule. If a report file already exists for the same Sunday bucket, do not silently overwrite. Open the existing report, list its top-3 findings, and explicitly state in chat what changed in the new data before regenerating. A second run on the same week without a delta is wasted work.
No devs / reqs in user-facing strings. All UI text — callouts, table headers, KPI labels, verdicts, badges — must say devices and requests. Internal variable / column / file names in scripts and JSON can stay short.
Do not create a separate Markdown summary of the report — the HTML is the deliverable.
Do not commit the report file. It lives in $env:USERPROFILE\android-oce-reports\ (outside the workspace) precisely so it can't be staged accidentally.
Output checklist
New oncall-wow-report-YYYY-MM-DD.html (where YYYY-MM-DD is the resolved curEnd — the end-date of the rolling 7-day window) exists at $env:USERPROFILE\android-oce-reports\ (NOT at repo root). If a file for this end-date already existed, the chat session explicitly stated what changed before regenerating.
All sections present and populated (incl. 🚚 Traffic Attribution — even if “None this week”)
60-day trend bucketing run on the full cross-product — {error_code, error_type} × {devices, requests} = 4 runs — union of regressions reported. Per-request retry storms (e.g. small device pool, exploding request count) are flagged on both axes. Source KQL spans the literal last 60 days ending today (no source-side partial-week filter); the partial current week is excluded from delta classification via bucket-trends.js --end=<startofweek(today)> --include-partial-end and charted as the final bar.
WoW-movers pass run (wow-movers.kql) for BOTH error_code and error_type. Its output rows are merged into the single 🔴 WoW regressions callout in Section 2 (sorted by curr-week devices descending), each row tagged NEW / 60d↑ / originator chip. No separate "emerging" callout. Every row carries throw-site, dominant message, originator, and a next step. If the WoW callout is empty (rare), render "None this week" rather than omit.
Both error-codes AND error-types WoW tables have Δ requests % and Δ devices % columns, the 60d sparkline, and a status pill. Any row crossing threshold on either metric is in the regression list.
Every WoW regression AND every 60d regression — for both error_code and error_type — has its own spike-attribution card with all 7 dimensions sliced. Cards are built from assets/templates/spike-card.html.
Every error_type regression card includes the 8th-dimension sub-code decomposition showing the top 3–5 contributing error_codes with their Δ vs prior week, and links to those sub-codes' own attribution cards.
Originator pre-check has been run for every broker-tagged card (error-message-and-location.kql). Throw site and top 3 error_message strings are populated from real data, not from the code map. AADSTS-prefixed messages are tagged eSTS, not Broker.
Every regression card's Code Attribution block populates Originator + Top throw site + Wrapper + Caller hot-spots + Underlying cause + Top error_messages + Likely PRs (with confidence/why-it's-the-suspect) + Next step (with named owner). For type cards, the wrapper field focuses on the type's catch-and-rethrow site (e.g. BaseException, ServiceException constructor). Shallow PR-only attribution is not acceptable.
Non-broker errors are explicitly tagged environmental / non-broker with confidence none — not invented broker PRs.
Traffic analysis covers totals, per-app, per-span, requests-per-device ratio (per error AND overall), and a sampling-change check.
Every material traffic shift (>10% on any segment, up or down) has a reasoning paragraph that names the dominant span/app/active-broker/broker-version, and either cites a causal PR (with confidence) — span removed/added, goAsync() refactor, sampling change, caller-side SDK release, ECS flight ramp — or explicitly says "no PR identified, suspect X" rather than leaving it unexplained.
Auth-only denominator used for all reliability %s, denominator caveat called out at top.
No \bdevs\b or \breqs\b in user-facing text. (Select-String -Pattern '\bdevs\b|\breqs\b' -CaseSensitive:$false returns 0.)
Sparklines rendered. Every .kpi tile in the Top-line health section has a data-spark array with 8–9 weekly values. Every row in the 60-day trend tables and both WoW tables (codes + types) has a data-trend mini-spark. Note the 60-day trend data-trend arrays now end on the current partial week (≈9 points incl. the in-progress bar), while the WoW-table sparklines keep 8 complete weeks. The validator's chart-coverage check passes (KPI coverage ≥1/2 of tiles, total elements ≥15). Past failure mode: the v7 body rebuild dropped all sparklines silently — see template-readme.md § "Sparklines are MANDATORY".
Code-attribution depth. Every .attr-card's Code attribution block uses the full 8-field <div class="origin-row"> structure (Originator / Top throw site / Wrapper / Caller hot-spots / Underlying cause / Top error_messages / Likely PRs / Next step) per assets/docs/code-attribution-template.md. A pr-list-only stub is not acceptable — the validator hard-fails this. Past failure mode (v7 third pass): all 10 cards shipped with PR-only stubs and lost the throw-site / wrapper / underlying-cause analysis.
No stale text from previous weeks. (Select-String -Pattern 'EXAMPLE CONTENT BELOW' returns 0 — that's the unfinished-section sentinel. The template no longer ships {{TOKEN}} placeholders since v2; if the file still contains any {{, that's also a leftover.)
