mathias

mathias pushed to feat/33-audio-upload-transcript at mathias/tapir

2026-10-02 06:01:33 +00:00

1a89eb3868 docs(adr-034): record the first blind check; bar ambiguity and the unanswered second condition

2e9d76dad0 docs(research): first blind check of two-engine reconcile on a real recording (aggregates only)

Compare 2 commits »

mathias opened issue mathias/tapir#43

2026-10-02 05:30:21 +00:00

Upload worker logs nothing until the job ends: no start, phase, engine or adjudication lines

mathias commented on issue mathias/tapir#42

2026-10-02 05:26:23 +00:00

Upload page: show progress, so the user can tell running from hung or crashed

Requirements from the product owner (2026-10-02), to be treated as part of this issue's acceptance:

Visual style. The indicator continues tapir's existing visual style. Reuse the…

mathias opened issue mathias/tapir#42

2026-10-02 05:24:09 +00:00

Upload page: show progress, so the user can tell running from hung or crashed

mathias pushed to feat/33-audio-upload-transcript at mathias/tapir

2026-10-02 05:13:10 +00:00

7f8e3677b3 docs(adr-034): gemma4-31b slowness explained; reopen the 26B-vs-31B reconciler choice

e3aa661cf2 docs(research): Low Power Mode confirmed as the cause of slow gemma4-31b (22.3 vs 7.9 tok/s decode)

Compare 2 commits »

mathias commented on pull request mathias/tapir#39

2026-10-01 21:28:25 +00:00

feat: upload audio, get a private Swedish transcript, discuss it (#33, ADR-033)

Scope added since the description above: two-engine transcription with a local reconciler (ADR-034, migration 019, internal/reconcile, transcribe.Ensemble). Off by default; the existing…

mathias commented on issue mathias/tapir#40

2026-10-01 21:28:12 +00:00

Add iguana/pianissimo as a second local transcription backend (middle tier, or run both)

Built: option B (run both, reconcile locally) — ADR-034, on PR #39

Chosen by the maintainer. Off by default; TAPIR_RECONCILE_BASE_MODEL=iguana/pianissimo + `TAPIR_RECONCILE_MODEL=iguana/…

mathias pushed to feat/33-audio-upload-transcript at mathias/tapir

2026-10-01 21:28:07 +00:00

7ae888a468 docs(claude): two-engine reconciliation, and that task check does not gate lint (#41)

b9e0dfa1c9 docs(homelab): two-engine mode, its settings and why every model must be local

8dbcb9a9eb docs(data-model): reconciliation column

26cc00c1cc test(acceptance): map the two-engine scenarios to their covering tests

36a80ac4f5 docs(use-cases): scenarios for two-engine reconciliation (ADR-034)

Compare 18 commits »

mathias opened issue mathias/tapir#41

2026-10-01 21:26:06 +00:00

Taskfile lint step can never fail: `golangci-lint run

mathias pushed to feat/33-audio-upload-transcript at mathias/tapir

2026-10-01 21:23:40 +00:00

6fb7a52980 feat(web): say what reconciliation did, and never claim a second opinion that was not there

d6ba11d288 test(web): the page says what reconciliation did, plainly, and never claims more (red)

ae4f9fd775 feat(store): persist and read back the reconciliation

e838f6bb6d feat(store): migration 019 -- uploaded_transcripts.reconciliation

5bd41077af test(store): the reconciliation survives the round trip, and its absence stays absent (red)

Compare 8 commits »

mathias pushed to feat/33-audio-upload-transcript at mathias/tapir

2026-10-01 21:20:41 +00:00

f25acebe4c feat(reconcile): merge two transcripts with a model choosing between disagreements

7f9e967ad0 test(reconcile): a verdict returned with an error is not used

0fead2e102 feat(domain): Reconciliation -- how a merged transcript came about (ADR-034)

a1e16fa658 test(reconcile): fixtures need two shared words between disputes

b20558ffa6 test(reconcile): merging two transcripts, and failing soft (red)

Compare 14 commits »

mathias pushed to feat/33-audio-upload-transcript at mathias/tapir

2026-10-01 21:15:14 +00:00

de9871c9c3 docs(adr): ADR-034 transcribe with two engines, reconcile with a local model

e182a6b07d docs(research): can a local model adjudicate two ASR transcripts?

Compare 2 commits »

mathias pushed tag v0.19.0 to mathias/hyperguild

2026-10-01 21:10:09 +00:00

mathias commented on issue mathias/gitea-mcp#64

2026-10-01 21:09:46 +00:00

tools/call is not audit-logged — only auth acceptance is (OWASP MCP08)

The Loki clause I left unverified earlier is now verified. `{namespace="gitea-mcp"}

mathias closed issue mathias/hyperguild#144

2026-10-01 21:09:45 +00:00

brain-mcp: one shared bearer for every client, no per-caller read scoping (OWASP MCP07/MCP10)

mathias commented on issue mathias/hyperguild#144

2026-10-01 21:09:44 +00:00

brain-mcp: one shared bearer for every client, no per-caller read scoping (OWASP MCP07/MCP10)

The last done-when clause is verified live. PR #147 deployed as ingestion:a6da933 (infra 36916a3), and the rollout completed. I then ran a live brain_query through the plugin MCP, and Loki…

mathias reopened issue mathias/hyperguild#144

2026-10-01 21:07:32 +00:00

brain-mcp: one shared bearer for every client, no per-caller read scoping (OWASP MCP07/MCP10)

mathias commented on issue mathias/hyperguild#144

2026-10-01 21:07:31 +00:00

brain-mcp: one shared bearer for every client, no per-caller read scoping (OWASP MCP07/MCP10)

pr-gatekeeper: PR #147 merged as a6da933. One structured mcp tool call slog record per tools/call, carrying tool, principal, origin, wing, hall, content_bytes, duration_ms and outcome (ok/error…

mathias commented on issue mathias/hyperguild#145

2026-10-01 21:07:30 +00:00

brain_write is an ungated persistent prompt-injection path into every future session (OWASP MCP06)

pr-gatekeeper: done, verified by execution. PR #147 merged as a6da933 after four review rounds.

Write side: brain_write, brain_update, brain_edit, brain_promote, capture (MCP and REST), MCP…

mathias opened issue mathias/hyperguild#151

2026-10-01 21:07:17 +00:00

provenance follow-ups: decode written_by on read; promote leaves empty dir on refusal; partial multi-page ingest