Root cause of the ~5,148 unpublished 8B/5B/SCM PASS records (driving the fix):
(1) parseRawData wrongly consumes a PASS/FAIL line as the step-response line for
non-DSCA families that omit the "0","0",v line (8B45/8B49/5B39/SCM5B33...) ->
drops the first Final-Test group -> measurement-count mismatch -> null render.
(2) Even parsed, the renderer has ONE hardcoded DATA_LINES['8B'] (RTD-shaped), so
models like 8B45 (frequency input, == DSCA45 structurally) get wrong param
names/specs. Same class as DSCA -> needs per-model templates.
Mined per-model templates from the Hoffman originals (published siblings) for all
136 mineable models via tools/mine-hoffman-dsca.py (family-agnostic extractor):
8b5bscm-templates.json = {accOut, accHeader, rows[name,spec], _srcSerial}.
Input-type split: 72 voltage / 18 temp / 12 current (accuracy already handled) +
10 frequency (8B45/5B45 — same unsolved freq-accuracy as DSCA45) + 24 7B/other.
Only 3 niche models (17 units) have no Hoffman original.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Backend (deployed live on AD2, service restarted, + repo copy resynced — it was
far behind the deployed server):
- /api/search: add whitelisted sort/dir (NULLS LAST) so sortable headers and the
"Latest uploads" preset work. web_status filter and POST /api/upload already
existed on the server; the stale repo copy now matches live.
Frontend (redesign prototype):
- "Latest uploads" preset (web_status=on + sort=api_uploaded_at desc) and
"Not yet published" (web_status=off) are now active presets.
- Push to Web (inspector) + Re-push (multi-select) wired to POST /api/upload
behind a confirm() gate; refresh WEB status after. Validated idempotently on a
published record (unchanged:1, errors:0).
- "Retested units" stays disabled — needs a retest flag in the pipeline (next).
tools/preview-proxy.py: forward POST so the publish buttons work in same-origin preview.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- fitCert: replace the flaky CSS `zoom` (Firefox support is recent/inconsistent)
with transform:scale() measured against the widest line (+ right margin and
font-load retries) so the cert always scales to fit the inspector with no
horizontal clip. Validated live on a narrow 5B cert (0.74x) and a wide DSCA45
cert (0.55x) against the real AD2 dataset.
- inspector Web field -> Published (green) / Not published (amber) chips.
- widen default inspector 480 -> 500px.
- tools/preview-proxy.py: serve the prototype AND reverse-proxy /api to the live
AD2 server so the cert iframe is same-origin during preview — styleCert/fitCert
read iframe.contentDocument, which silently no-ops when the iframe is loaded
cross-origin straight from AD2 (why the fit looked broken in earlier previews).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The DSCA33/DSCA45 main spec files lost in the cryptolocker wipe are recoverable:
the original software published correct certs to the Hoffman product API before
the wipe and our null-skipping renderer never overwrote them. Mine per-model
Final-Test templates (names + specs + verbatim accuracy headers) straight from
those originals instead of requesting spec files from Dataforth/John.
- dsca33-45-templates.json: 56 models (DSCA33 34/35, DSCA45 22/23); only
DSCA33-1948 + DSCA45-1746 (24 units) lack an original.
- mine-hoffman-dsca.py: the re-runnable miner.
- DSCA33-45-HOFFMAN-RECOVERY handoff for the AD2 session (incl. the gate:
validate each render vs its Hoffman original before enabling live rendering).
- memories: Hoffman recovery (supersedes the spec-gap "need John" note) and the
AD2 SSH MTU-blackhole root cause/fix; errorlog entries (syncro jq, ssh correction).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Vanilla single-file (no build/CDN) command-center redesign of the testdatadb search
UI: omni-search with serial/model/text routing + auto-select fast path, dense
monospace results table with PASS/FAIL pills + web-published indicator, persistent
split-pane datasheet inspector (iframe to /api/datasheet), left filter rail
(result/date/model/station/log), server pagination, CSV export, URL state, keyboard
nav (/ ↑↓ ↵ Esc), clinical light theme. Hits the existing API; deployed to AD2 as
public/index.redesign.html (preview at :3000/index.redesign.html). Synthesized from
Grok + Gemini concepts (both converged on this command-center design).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Remote SSH/VPN to AD2 keeps flapping; hand the remaining datasheet fixes to the
local AD2 session. Includes the per-subtype approach (DSCA_TEMPLATES from staged
originals — STAGE 1 done, dsca-templates.json on AD2 = 126 models), the render-wiring
+ per-subtype byte-validation gate, Fix 5 (379 backfill via legacy_cert_text), the
discipline (backup/save-state/validate-before-publish), and the derive-dsca-templates
tool. Ref ticket #32441.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
While using the new 3-retry gemini path for live VPN research, two bugs surfaced:
- emit_or_fail checked auth_failed INSIDE the retry loop; a benign mid-run token-refresh line
matched the over-broad auth regex (bare login|credential|authenticat|oauth|401) and aborted the
retries with a false "auth error" - even though `gemini -p` auth tested fine. Moved auth-classify
to AFTER the retries (it only picks the final error message now) and tightened auth_failed to real
signatures (invalid_grant, not authenticated, login with google, token expired, ...).
- Added quota_exhausted() + a QUOTA FALLBACK: the pinned strong model (gemini-3.1-pro-preview) hit
"exhausted your capacity on this model" mid-session; emit_or_fail now retries once on the default
(lighter) model by stripping -m (separate quota). Validated: capped pro run -> fell back -> 2.9KB answer.
CT_THOUGHTS Thought 2 Resolution updated with both. (Search-bot reliability hardening continues.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Mike's must-fix. Diagnosed from RAW output of failing queries (not guessed):
- grok xsearch = TIMEOUT: grok-4.20-multi-agent web_search runs past budget on multi-part queries
(286s/280s, rc=124, still searching - 183 thoughts, only progress-noise text); buffered json => total loss.
- gemini search = INTERMITTENT empty turn (a clean re-run gave a real 2.6KB answer in 122s); the wrapper
retried only once, so two empties in a row failed spuriously.
Fixes:
- ask-gemini.sh emit_or_fail: retry up to 3x with 3s/6s backoff (was 1).
- ask-grok.sh xsearch: --output-format streaming-json (salvage partials) + AUTO-FALLBACK to
ask-gemini.sh search when grok doesn't finish (rc!=0 or empty). Validated e2e: grok timed out
(rc=124) -> fell back -> gemini returned a real sourced answer (UniFi Teleport invite-link API).
grok's own multi-agent timeout is an xAI-side limitation; the fallback makes xsearch reliable regardless.
Docs: grok SKILL.md xsearch row + CT_THOUGHTS Thought 2 Resolution.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
