TASK-IMP-091: regen_backlog emits every status and recomputes Totals from frontmatter

Summary

The function every insert contract cites as the byte authority cannot regenerate a truthful index: its ACTIVE-status filter drops done rows entirely - the proven root cause of the fourteen-task drift TASK-IMP-086 repaired. Make regen_backlog emit one row per task folder for every status, recompute the repo-wide Totals line from a frontmatter tally, and gate it with a parity test that regenerates today's corpus and byte-compares the improvement section against the committed file.

Problem

TASK-IMP-086's mandated trial recorded the failure precisely: regen deleted the three committed done rows, emitted nothing for 068-081 (all done), and left a Totals line three short of the corpus. An index regenerator that silently drops terminal statuses turns the audit trail into fiction the moment tasks ship.

Proposed Solution

Remove the ACTIVE filter from row emission (the index lists the whole corpus; the queue's eligibility logic stays elsewhere), keep grammar and stem ordering byte-identical to the committed convention, and compute Totals from a per-status frontmatter tally at regen time. A new suite regenerates against the live corpus into a scratch copy and asserts (a) byte-parity of the improvement section with the committed file and (b) Totals equal to the tally.

Alternatives Considered

• Re-point the byte-authority citation at backlog-mutate.mjs and retire regen. Rejected for now: full-section regeneration from frontmatter remains the recovery tool for exactly the drift class 086 hit; fixing it is one filter.
• Emit done rows into a separate archive section. Rejected: the committed convention (086's repair, reviewed and merged) lists them in place with the status tag.

Success Metrics

• Primary: regen on today's corpus reproduces the committed improvement section byte-for-byte and a true Totals line, asserted by the suite on every run. Baseline: regen output deletes 17 rows and undercounts Totals by 3 (recorded trial). Deadline: final acceptance.
• Guardrail: a fixture corpus with one task per status yields one row per status (the filter cannot silently return).

Scope

In scope: the emission filter, the Totals computation, the parity suite.

Out of scope / Non-Goals

• Other sections' drift repair (regen fixes the mechanism; running it repo-wide is an operator choice).
• The migrate/adopt phases of the script (row regeneration only).

Dependencies

• None; cone is the python script plus a new scripts/tests suite (disjoint from all batch members).

AI Authorship Disclosure

• Tools used: Claude (Fable 5) running the CyberOS task-author skill inside Cowork.
• Scope: spec drafted from the recorded TASK-IMP-086 trial evidence; implementation under ship-tasks supervision.
• Human review: batch-3 PLAN approved 2026-07-16; both HITL gates are recorded human verdicts.

1. Description (normative)

• 1.1 regen_backlog MUST emit exactly one row per task folder for EVERY frontmatter status (the ACTIVE filter is removed from emission), preserving the committed row grammar and stem-ascending order.
• 1.2 The Totals line MUST be recomputed from a per-status tally of the corpus frontmatter at regen time.
• 1.3 Regenerating today's corpus MUST reproduce the committed improvement section byte-for-byte (rows and header), proven in the suite against a scratch copy.
• 1.4 A fixture corpus containing one task in each lifecycle status MUST yield one row per status.
• 1.5 The suite MUST land at scripts/tests/test_regen_backlog.sh, discovered by the run_all glob.

2. Acceptance criteria

• [ ] AC 1 (traces_to: #1.1, #1.3) - live-corpus regen is byte-identical to the committed section - test: scripts/tests/test_regen_backlog.sh::t01_live_corpus_parity
• [ ] AC 2 (traces_to: #1.2) - Totals equals the frontmatter tally - test: scripts/tests/test_regen_backlog.sh::t02_totals_true
• [ ] AC 3 (traces_to: #1.4) - per-status fixture emits every row - test: scripts/tests/test_regen_backlog.sh::t03_every_status_emitted
• [ ] AC 4 (traces_to: #1.5) - suite discovered by run_all - verify: suite listed in the runner output (ops check in the gate log; glob discovery is the runner's contract).

3. Edge cases

• Titles containing the row separator or other task ids (081's title quotes TASK-IMP-080): emitted verbatim; parity is byte-level so any mangling fails t01.
• A folder with unparseable frontmatter: regen MUST fail loudly naming the file, never emit a guessed row (asserted inside t03's fixture set).
• Sections other than improvement: untouched by this task's assertions; the mechanism fix applies wherever the script is pointed.
• Security-class: none - a read-compute-write script over tracked markdown, gated by byte-parity.