Roadmap v11 — evidence-led forecast accuracy closeout¶
Status dated: 2026-09-27
All 16 roadmap work items are closed. The closeout is inconclusive for forecast accuracy: the required historical point-in-time corpus remains unavailable, recent prospective/dashboard evidence is not an admissible canonical historical evaluation, and no candidate was selected. This outcome supports no positive or negative accuracy claim and authorizes no production model, input, interval, or weighting change.
Workstream outcomes¶
| Workstream | Issues | Outcome and evidence |
|---|---|---|
| Evidence continuity and admissible data | #407, #408, #409, #410 | The independent backup and prospective ledger recovery/continuity items are closed. The canonical corpus gate remains blocked: the audit documents 0/32,136 target hours and 0/4,320 warm-up hours, with no admissible versioned BTC/USD archive acquired. The frozen scoring contract is recorded in #410. |
| Production diagnosis | #411 | Production and registered baselines were captured, and the frozen replay-readiness audit was run. The audit is blocked: its last verified snapshot had 48 timestamp-eligible rows, all legacy/unversioned and excluded, with zero confirmatory origins or matured pairs. Metrics are null; this is not a scored replay. See the frozen cohort record and its issue report. |
| Conditional candidate evaluation and selection | #412, #413, #414, #415, #416, #417 | All six dispositions are blocked/not attempted because #411 did not provide a powered diagnosis. No candidates were scored, no winner was selected, and D3 was not launched. See #412, #413, #414, #415, #416, and #417. |
| Mature-cohort and dashboard evidence integrity | #419, #420, #421, #422, #423 | Frozen mature-cohort accounting and dashboard coverage, segment, rolling-window, and volatility-attribution reporting were completed. The audit noted 200 exact matured outcomes out of 208 expected pairs, with 8 right-edge pairs naturally pending; this is not a candidate skill result. |
The issue-level discussions, closed states, and linked artifacts are the detailed outcome record for each work item. The supplemental audit in the #418 issue discussion records a separate short-history dashboard diagnostic: all 30-day, 90-day, and all-history windows contained the same approximately 22-day sample, and the minimum effective-block gate was not met. Its paired deltas and interval are not a canonical skill estimate and do not override the blocked #411 replay.
Frozen research evidence¶
The checked-in machine-readable records provide immutable versions and digests for the blocked research decisions:
| Evidence | Report/version | SHA-256 or protocol digest | Result |
|---|---|---|---|
| #397 canonical corpus audit | audit artifact | e2bf0c671a5a4327c6852d02614a7d2e403ed12f8fe7c4362336c5e87d44a0c7 |
0/32,136 target and 0/4,320 warm-up hours |
| #409 corpus comparison | artifact v1 | f1991ec340479c9cb2e617f36892123c70428a9522d5620203aff3c540e1c266 |
Five channels reviewed; none admitted; no provider contacted or data acquired |
| #410 frozen scoring protocol | v11-prospective-1.0.1; file digest 1b8c6705d53882a66c0ab1277cf2a63c845d00c2c7d0cc978c0c75d8121efd18; protocol digest 4da936d505df9a15cbd073ee85ee0e4ed2db483e6058f6b81ce163f4724bb4f0 |
Digests identify the frozen pre-score contract | Protocol gate passed; it does not assert scoring occurred |
| #411 replay readiness | schema v1; file digest 7561c14b141b69bd1c05f0926434a258e8baef410c2fd62613e2e1237288bc71 |
Snapshot database b102f14db535264ecc2474a3b338c929abea4c39616f233a5701611c9300f57f; cohort 462287eb433a27c16e4f25360d9ee40e5dda2fabdced13e5390859a055f7bbb8; audit code 79a95fc3debd4134ab68cfc55795007a340f15a6eff2ef7212a875a9846c28fc |
48 eligible timestamps, all legacy-excluded; zero confirmatory origins/pairs; metrics null |
| #417 D3 disposition | artifact v1; file digest 6c7356a79d1295c6332ba40bc1cace15f478016ccbef4868b6d48f81b986c8dd |
Candidate/champion and metrics null | D3 not launched |
The artifacts' issue-specific records preserve the commands, schemas, and reopening gates. No licensed corpus, private runtime database, or forecast-score data is included in the repository.
Backup incident follow-up¶
The later backup incident #440 was a recurrence of the already tracked #407 failure mode. The independent monitor initially reported a verified but stale archive. Follow-up verification and a scratch-only disaster-recovery drill passed:
| Check | Run and result |
|---|---|
| Independent backup monitor | Run 36346607970, passed at 2026-09-27 20:04 UTC. Report v1: schema 6, 147 origins, 4,092 predictions, 3,966 matured predictions, SHA-256 df4a001580ed480f6d2e0e685b14c4b4fedae1e449e89ca77e9a4c5822cc3eeb, backup age 6,317 seconds (below the 2-hour gate). The independent shadow backup also passed at schema 7. |
| Disaster-recovery drill | Run 36347546910, passed at 2026-09-27 20:19 UTC. Report v1 verified SQLite integrity and schema 6 in scratch storage, matched manifest counts/hash, and recorded restore receipt at 20:19:38 UTC. The shadow database also restored and verified at schema 7. |
The drill used the repository workflow's scratch-only recovery path; it did not replace production history. Report artifacts are retained on the linked Actions runs. The failed run report is retained at run 36335817092.
Reproduction and validation¶
The canonical corpus audit command and required immutable input conventions are
documented in the corpus acquisition audit.
The operational backup checks are reproducible through the manual dispatches
of independent-history-backup-monitor.yml and
disaster-recovery-drill.yml; the linked runs above record their exact commit,
commands, reports, and check results. No licensed market data, private runtime
database, backup archive, or generated workflow report is committed here.
Final disposition¶
The roadmap work is complete, but the evidence chain does not meet the frozen requirements for an accuracy-improvement claim. The canonical source remains blocked; available prospective evidence is limited and does not meet the historical skill gate; and no candidate passed selection and confirmation. Retain the production policy. Reopen evaluation only after its stated corpus, sample-size, provenance, maturity, and inference gates are satisfied.