Merge remote-tracking branch 'box/master' into enforced-proving

docs/design/app-audit-2026-10-08.md

@@ -358,4 +358,5 @@ Recorded as the founder's call, read through the shipper; they override section
 | F3 | The four interface words on the wallet and the window | included, executed, proven, finalised (docs/spec/finality-guarantees.md section 9); a locked block is finalised even in a pause, the pause sits on the status strip, no block reads "finality not active" | the window's block inspector and the scene's words (cards-201); the wallet's History row (`app/igneum-wallet/ui/app.js` stateWord, branch wallet-words-20 on master) |
 | F4 | One positioning line | the hero keeps it; `#s-positioning` under Updates is gone | cards-201 |
 | F5 | The watchdog's words while the node serves no epoch state stream | "waiting for the node's execution state for this epoch", never "the worker did not load its program" | `app/igneum-app/src/watchdog.rs` on cards-201 |
+| F6 | A node with no peers holds the worker, never a worker fault (19:25 BST on the mini: the worker idled for lack of templates and the watchdog blamed it, "did not come back within 180 s of the miner restarting it") | whenever the node reports zero peers or no template for the window, every watchdog clock holds (the restart grace included) and the card reads "no block templates: the node has no peers" (or "…answers none for the window"); the node line reads "Node: no peers, dialling the seeds" | `app/igneum-app/src/watchdog.rs` (`tick(now, node_ready, no_peers)`, `held_by_node`), `engine.rs`, `ui/app.js` on branch peers-202 for v2.0.2 |
 

docs/plans/igneum-2.0-gap-table.md

@@ -0,0 +1,25 @@
+# Igneum 2.0 gap table (the Test and Acceptance Standard's profiles against today's values)
+
+Built by the coordinator for main's 20:30 ask on 8 October 2026 (19:2x UK). One row per profile: the cases it governs (from the registry on master), today's value in the team's own words, the verdict today against the profile as approved, and the clock of the first confirmatory run under the standard's procedure. "Team-run" means a run with evidence before the procedure; nothing reads PASS under the standard until the steward's test-record writes it.
+
+| Profile | Title | Cases governed | Today's value | Today against the profile | First confirmatory run |
+|---|---|---|---|---|---|
+| P00 | Frozen scope and approval | 14: GOV-01 GOV-02 GOV-03 GOV-04 GOV-06 GOV-08 POW-07 POW-08 ROT-03 ROT-04 VER-04 OPS-08 LEAD-07 LEAD-08 | The standard approved as proposed at 18:2x (P13 deferred); the registry on master with an owner on every case; F0 (the release manifest) not yet signed: the steward's test-map (37 of 109 automated cases mapped to 17 cells) is on master at e704b920; the genesis identity frozen (devnet-4 4cdcc488, digest be5f4068, chain id 4465); the workload W, the fault model and the claim inventory open | NOT RUN (BLOCKED fields: W, the fault model's thresholds in F0) | F0 signed by the steward: 23:30 tonight (GOV-01), the claim inventory with it |
+| P01 | Correctness and negative-test depth | 69: GOV-05 GPU-04 GPU-05 GPU-07 POW-01 POW-02 POW-05 POW-06 POW-07 ADV-01 ROT-01 ROT-02 ROT-03 ROT-04 ROT-05 ROT-06 EVM-01 EVM-02 EVM-03 EVM-04 EVM-05 EVM-06 EVM-07 EVM-08 ZKP-01 ZKP-02 ZKP-03 ZKP-04 ZKP-05 ZKP-06 ZKP-07 ZKP-08 CAP-02 CAP-05 CAP-07 CAP-08 INC-01 INC-02 INC-05 INC-06 INC-07 FIN-01 FIN-02 FIN-03 FIN-04 FIN-05 FIN-06 FIN-07 FIN-08 VER-01 VER-02 VER-03 VER-04 VER-05 VER-06 VER-07 VER-08 OPS-02 OPS-03 OPS-04 OPS-05 OPS-06 UX-02 UX-03 UX-04 UX-05 UX-06 COM-04 LEAD-06 | P01 part B done (19:07): three parsers, 10,000 seeded malformed cases each, 0 panics, every case under the bound; the pack reader 10,000 mutated files, 911 refused, 0 crashes; the Rust parsers 10,000 each, one real fault found and fixed (bind::unhex on a multi-byte character); the one-million full-hash vectors per backend generating on box 2 (four packs by 20:20), the CUDA backend on the P01 pods from 21:45, all backends by 23:30 (AMD and Arc on PC 1 and PC 2); EVM-02 to 06 and VER-03 to 06, 08 PASS on devnet-4 at 18:45 (team-run, not yet under the procedure); VER-01 and VER-02 FAIL by design and disclosed | RUNNING; 10 PASS team-run, 2 FAIL by design (VER-01 voter table as trust anchor, VER-02 no SP1 verifier in the client), no invalid acceptance observed | P01 part A (the vectors) first backend 21:45, all 23:30; the parser half PASS tonight on the steward's write-back by 19:25 |
+| P02 | Hardware coverage and reproducibility | 12: GOV-03 GPU-01 GPU-02 GPU-03 GPU-04 GPU-05 GPU-06 GPU-07 GPU-08 CAP-01 UX-08 LEAD-02 | The cohort tonight: NVIDIA 2080 Ti, 3070 8 GB, 4060 8 GB, 4070 Ti 12 GB read at stock (fingerprints as expected); 3060 and 4070 re-rented on 580 drivers, 1080 Ti booting, 5090/4090/3090 on P01; AMD: no rentable retail card on any provider, PC 1's RX 7600 and the morning's 9070 XT only; Apple: the Macs; Intel Arc B580 on PC 2; three unaffiliated operators and the seven-day soak not started | NOT RUN against the profile (3 AMD configurations impossible tonight; 2 Apple pending; the soak not started); the NVIDIA rows RUNNING | the full NVIDIA cohort rows 23:30; the AMD gap stated to the founder (a second AMD card needs a purchase or a loan); the soak starts on the v2.0.1 entry |
+| P03 | Candidate improvement and honest-card budget | 8: GPU-02 GPU-03 POW-03 POW-04 POW-07 POW-08 ROT-02 ROT-07 | Layer 8 off against its pair at stock: +1.7 and +1.9 percent per hash on the 5090, +2.5 and +1.0 on the 4090 (within one run's noise, under 5 percent); the knee rows at the 1,300 lock on PC 1 by 20:30 (RunPod containers refuse the lock); the genesis dataset size is the baseline, not a P03 cost (main's ruling) | RUNNING; PASS at stock, the lock pending | 20:30 (the knee rows); the D1 object decision on them |
+| P04 | Scoped specialist-competition target | 14: POW-03 POW-04 POW-05 ADV-01 ADV-02 ADV-03 ADV-04 ADV-05 ADV-06 ADV-07 ADV-08 ROT-07 ROT-08 LEAD-03 | Placed rows today: the multi-family adversary at hit 0.50 reads 1.85x same node and 2.27x a node ahead without layer 8 (1.93x and 2.40x with it); the bracket 1.5x to 2.1x same-node is a FAIL to work against; the 32-lane placed full core (adv-f, 393,000 cells) lands 21:00, the genesis comparator (adv-g) 21:30 | FAIL (R_E above 1.5 same-node on every placed row today) | 21:00 (ADV-01/03/04 placed 32-lane), 21:30 (ADV-07, ADV-02); the chain's value stays FAIL until a change moves the placed row |
+| P05 | Measurement and inference protocol | 0:  | Pre-registration exists for ECO-05 (the register frozen before results) and the census harness (sub-version 3); the standard-wide pre-registration (metrics, cohorts, holdout seeds, run order) is not yet a document | NOT RUN (no case bound to P05 in the registry; the protocol document owed) | the steward writes the protocol into the harness map by 23:30; first use on the P01 confirmatory runs |
+| P06 | Memory and support policy | 4: GPU-05 ROT-06 ECO-07 CAP-02 | Measured today: the 12 GB tier mines and proves together at 4 GiB (12.0 of 12.3 GB on the 3060) and loses co-residence at 5.5 GiB (13.6 GB); the 8 GB tier refuses 5.5 GiB by VRAM; the component memory budget not yet published | RUNNING (GPU-05 size cells on the 8 and 12 GB pods tonight); the budget NOT RUN | GPU-05 rows 23:30; the budget document with the dataset decision at 20:00 |
+| P07 | Proof service capacity and fairness | 15: ZKP-08 CAP-01 CAP-02 CAP-03 CAP-04 CAP-05 CAP-06 CAP-07 CAP-08 INC-03 INC-04 INC-05 OPS-01 COM-01 COM-04 | W not frozen; the proving benchmark per stage (CAP-01 to CAP-08) starts on devnet-4's first paid segment after the first lock (the window fills about 19:08); the 72-hour run at W not started | NOT RUN (W BLOCKED until frozen) | CAP rows by 15:00 tomorrow as set; W frozen by the founder's word (asked with the F0 fields) |
+| P08 | Fault assumptions and recovery | 25: ROT-01 ROT-02 ROT-03 ROT-04 ROT-05 EVM-08 ZKP-08 CAP-07 INC-04 INC-06 INC-08 FIN-01 FIN-02 FIN-03 FIN-04 FIN-05 FIN-06 FIN-07 FIN-08 VER-05 VER-06 OPS-01 OPS-02 OPS-04 OPS-08 | Rule v4 (the anchored table, majority-continuity recovery) in the spec at 4b423dad8 and in the node at acd9d067, not in any live path (devnet-4 has rule v4 at never); the harness: the recovery run as designed (side A 51.26 percent recovered at index 11, side B paused, 0 conflicts, 0 disagreement), the pause-only fault found and fixed; the real-network partition case open | RUNNING (FIN-07, FIN-08 harness rows; ROT rows on 417c4a57 green); the live-network case NOT RUN | the HEAL=400 reruns on the successor pair tonight (about 20:10 tables); the first live partition drill after the roll |
+| P09 | Security and bounded resource requirements | 30: GOV-07 POW-06 EVM-01 EVM-03 EVM-04 EVM-05 EVM-06 EVM-07 EVM-08 ZKP-01 ZKP-02 ZKP-06 ZKP-07 INC-05 VER-01 VER-02 VER-03 VER-04 VER-05 VER-06 VER-07 VER-08 OPS-02 OPS-03 OPS-05 OPS-06 OPS-07 UX-05 UX-07 COM-08 | The internal attack pass F1 to F10, the binding review (A05 closed at ad5a54c5), the review's nine findings (five fixed by construction on class-v6, A03 and A05 closed, A01 by construction, A09 not applicable); key succession green on 291ee6ae; no external scope by the in-house rule; the disclosure prize the only outside check | RUNNING (zero unresolved critical findings known; the resource bounds measured on the parsers) | the review verdicts by 22:30, the D1 object 23:30; the prize page 23:00 |
+| P10 | Ordinary-operator product targets | 11: GPU-06 GPU-07 OPS-07 UX-01 UX-02 UX-03 UX-04 UX-06 UX-07 UX-08 LEAD-02 | UX-01: the Windows 2.0.0 entry live 18:44, both PCs' MINING-ON read-back by 19:25; the Mac fresh install on the mini at 11.0 MH/s (18:13); 30 unaffiliated participants not recruited | RUNNING (UX-01 to UX-03, UX-07 tonight on the canary); the 30-participant study NOT RUN | UX rows 21:30 to 22:00; the study is a founder decision (recruitment) |
+| P11 | No-founder exercise and independence | 4: INC-08 OPS-01 OPS-08 LEAD-05 | 92 devnet boxes are the fleet's own; 27 distinct miner keys in ten minutes at 18:40 are the team's; no unaffiliated operator yet | NOT RUN | after launch; the builder programme's first outside node |
+| P12 | Five-year coexistence envelope | 24: ADV-04 ADV-06 ADV-07 ROT-08 ECO-01 ECO-02 ECO-03 ECO-04 ECO-05 ECO-06 ECO-07 ECO-08 CAP-06 INC-01 INC-02 INC-03 INC-04 INC-06 INC-07 INC-08 UX-03 COM-03 LEAD-03 LEAD-04 | ECO-05 run on a register frozen before results, R corrected to USD 31.65 M: FAIL (one of sixteen worlds sustainable under the two-vendor rule at 10R and USD 0.03; in it the GDDR7 machine's best core cell 1.46 to 1.70, cohort median 2.9 to 3.2; p90 fails everywhere); served as FAIL | FAIL | the AMD and Intel energy rows (GPU-03) 23:30 move the honest side's efficiency; the cube re-run on them tomorrow |
+| P13 | Maintenance continuity | 1: COM-07 | DEFERRED by the founder | DEFERRED | none |
+| P14 | Genuine commercial and developer proof | 10: CAP-08 INC-02 INC-07 COM-01 COM-02 COM-03 COM-04 COM-05 COM-06 COM-08 | no paying buyer, no external jobs; the reference apps and the DEX serve on devnet-4 | NOT RUN | after launch |
+| P15 | Comparative contention threshold | 5: COM-05 LEAD-01 LEAD-02 LEAD-03 LEAD-07 | no peer set pre-registered | NOT RUN | the steward's F0 names the three peers by 23:30 (LEAD-01) |
+| P16 | Observed durability and claim freshness | 5: LEAD-04 LEAD-05 LEAD-06 LEAD-07 LEAD-08 | devnet-4 born 17:14 today; 0 elapsed days on the release family | NOT RUN | 90 days from the final compatible release |
+
+Counts at 19:2x UK: P12 FAIL; P04 FAIL; P03 RUNNING (PASS at stock, the lock at 20:30); P01, P06, P08, P09, P10 RUNNING; P00, P02, P05, P07, P11, P14, P15, P16 NOT RUN; P13 DEFERRED. Four fields BLOCKED confirmatory runs until main's word at 19:3x UK, now given and written into F0 by the steward: (1) the workload W (P07) = the pilot plan's shape (one SP1 program pinned by vk, the shard fixture's size class, compressed proofs, one job an hour as the floor and the fleet's measured rate as the ceiling, requests/hour declared from the pipeline record's day); (2) the fault model and thresholds (P00, P08) = the finality spec's own (two thirds of the sliding table, the anchored table under rule v4, the recovery variant as the founder's default, the 40/40/20 fixture, the expiry cases as the spec lists them); (3) the 30-participant study (P10) = NOT RUN, "unaffiliated participants not yet recruited", the row kept open, never a staff run; (4) the second and third AMD configurations (P02) = the 9070 XT and the RX 7600 on PC 1 as the two retail AMD cells tonight, rented AMD cells added when any provider lists one, the row PARTIAL until then.