Files
BFM-decomp/config/kit_coverage_map.tsv

9.1 KiB

1kindkeydispositionnote
2ruleR2ENVall-in-WSL, one ext4 clone, loopback MCP — the source project's machine
3ruleR3PAtooling under tools/ is ProjectArchitect's layout convention (its templates carry it)
4ruleR4PAplan mode every phase = ProjectArchitect's gate 1
5ruleR5PAno attribution trailers — ProjectArchitect's SETUP/commit rule
6ruleR6PAthe agent commits, the owner pushes — ProjectArchitect's git posture (never push)
7ruleR7PAthe effort-map check before each phase plan — ProjectArchitect's effort protocol; the kit adds the decomp rows
8ruleR8PAcommit cadence — superseded in the source project by per-task commits (R42, cited); ProjectArchitect's own
9ruleR9FOLDED:G6+SEED:mcp-reconnect-after-restart.mdverify the oracle database persisted after a save — inside G6 (mirror + verify) and the reconnect seed
10ruleR11ENVthe Windows-native emulator bridged over a web API — the source project's machine
11ruleR12KIT:templates/run-README.mdno /tmp; project-local scratch under .run/ with dated tracked exceptions — the kit's run-README template + the firewall's scratch block
12ruleR14DK-68a summarised signal is a claim, not ground truth — realised as a kernel by the coverage pass
13ruleR18PAthe plain-English recap at every session/phase end — ProjectArchitect's PhaseEnd template
14ruleR19PAthe worklog archived under logs/, out of the load order — ProjectArchitect's close step (the kit's Step 10 does the same)
15ruleR20FOLDED:G18the private-repo backup policy, amended by R78 (cited) — G18's text carries the backup home
16ruleR21PAkeep the ops reference current in the same change — ProjectArchitect's ops-setup contract; the kit adds the decomp rows
17ruleR23ENVstop the Ghidra MCP before a checkpoint commit; the database churn is never staged — Ghidra-specific; the firewall template ignores the database directory
18ruleR25PAthe recap stored in the PhaseEnd — ProjectArchitect's PhaseEnd template
19ruleR26PAprompt for breadth (multi-agent) on breadth-shaped work — ProjectArchitect's effort protocol
20ruleR27PAevery effort transition prompted and waited for — ProjectArchitect's effort protocol
21ruleR28PAbuild the harness task list right after plan approval — ProjectArchitect's phase start
22ruleR30PAcapture context-dependent knowledge during the producing session — ProjectArchitect X5 (cited by the methodology)
23ruleR31PAthe decision log's WHY entries — ProjectArchitect's decision log; the kit's record ships the source project's
24ruleR52DK-66a blanket committer must not adopt a collapsed file — realised as a kernel by the coverage pass
25ruleR70DK-66a ledger's tie-break is part of the instrument — realised as a kernel by the coverage pass
26ruleR79DK-22pkill -f never with a literal the shell's own command line contains — DK-22's second wound
27ruleR80DK-66a widened checker classifies a missing promised page as PENDING — realised as a kernel by the coverage pass
28accelerator#12DK-25byte-locate the residual before blaming the draft — DK-25 is this entry under its own title
29accelerator#13DK-31census what is already drafted before drawing — DK-31 is this entry under its own title
30accelerator#14DK-26the cheap probe must call the real planner — DK-26 is this entry
31accelerator#18DK-20a claim derived from bytes is not verified by a compiler — DK-20 (a claim names the compilation it survived)
32accelerator#20DK-12unit boundaries at the jump-table spans at segmentation time — DK-12 (the segmentation law)
33acceleratorS77 (2)DK-19+G25a refusal-check measured against what the real pipeline does — DK-19 + G25 (negative-control every refusal-check)
34acceleratorS77 (3)G38an exclusion reason is a claim about the tooling on the day — G38 (draw-time bankability; re-probe the exclude list)
35acceleratorP32 T3 (2)DK-35a shared scratch directory is a shared blast radius — DK-35 (per-item work directories)
36acceleratorP32 T3 (4)COOKBOOKthe rematerialisation slider is a computation from the dump's counts — an idiom-level lever; lives in the cookbook corpus
37acceleratorP32 T3 (11)DK-44attribute the residual to a pass with a dump line — DK-44 is this entry
38acceleratorP32 T3 (14)DK-46the similarity score is not a shape oracle; grep the constants, objdump the window — DK-46
39acceleratorP32 T3 (15)DK-45re-run the producer census with the combiner's bookkeeping in view — DK-45 (a producer census before a spelling sweep)
40acceleratorP33.5 S91 (2)DK-57a typed figure against a derived one — DK-57 (numbers generated, never typed)
41acceleratorP33.5 S91 (3)DK-61a dry-run judge is an instrument; negative-control it — DK-61 (a known-true case before reading any instrument's output)
42acceleratorP33.5 S91 (4)G66a tool without a dictionary row is invisible — G66 (consult the tool dictionary first)
43ruleR84DK-67the ignore file's directory-form wall — DK-67 IS this rule
44ruleR85DK-57two DERIVED figures per check, never a typed one — DK-57 (numbers generated, never typed) carries it; G58 for documents
45ruleR86DK-66a dry-run judge is an instrument, negative-controlled before a FAIL is believed — DK-66 (the instrument's own parts)
46ruleR87G66a tool ships with its dictionary row or the health check fails — G66 (the tool dictionary)
47ruleR88DK-80re-run the link checker after ANY untrack or archive move — DK-80 (checks that pass vacuously before an untracking)
48ruleR89KIT:SETUP.mdan unattended installer's manifest is a derivable SET — the kit's installer states it that way
49ruleR90KIT:templates/firewall.txttranscripts kept in the repository need the firewall's class-7 line — the kit's firewall template carries it
50ruleR91RECORDa foreign project's path is named in prose, never as a backticked citation — a rule of this repository's link checker, not portable
51ruleR92DK-69a distillation ships with a coverage check against the populations it claims to distil — DK-69 (blind spots by construction)
52ruleR94DK-66a snapshot instrument never .strip()s a porcelain listing — one of the judge's own false flags, DK-66
53ruleR95DK-65types are a banking lever and a width lever — DK-65 (and G62's bank-time clause)
54ruleR96DK-81a scratch prune is an instrument change: every tool that writes under a pruned tree is re-run — DK-81 fact (4) IS this rule
55ruleR97G29green is read from a check's EXIT CODE with pipefail, never its last line — G29 (a soft error inside a success envelope is that error; provenance R49 + R53) carries it; the P34 masked-tail commit is its instance
56ruleR98DK-81a third-party hand-off file is proven through that party's own toolchain and the proving tool writes the paste — DK-81 fact (1) IS this rule
57ruleR99DK-81the public tree's contents decided before the flip; a retired document gets its Archive-index row and is deleted in the same commit — DK-81 fact (2) IS this rule
58acceleratorP36 S99DK-61the search harness whose target was a disassembly — a known-true case (a body the tree already matches) read before any verdict; DK-61
59acceleratorP36 S105DK-69a generator's refusals are an unaudited population — the tests that pass by construction; a negative-control corpus per generator; DK-69
60ruleR100DK-64one source per shared function, a duplicate a defect the health chain asserts — the dedup shape of DK-64 (readability at bank time) and the S1 gate; RECORD carries the P35 PhaseEnd
61ruleR101DK-59every task-advancing commit carries its log line and the checkpoint headline — DK-59 (the checkpoint is written to be replayed): a checkpoint older than HEAD is a dead session's
62ruleR102DK-66a tool restores from its own snapshot, never git checkout on a tree it did not commit — DK-66 (the restore path is part of the instrument)
63ruleR103DK-69a failure-cause extractor is negative-controlled against the compiler's real message forms — DK-69 (the tests that pass by construction)
64ruleR104DK-9build the disagreeing oracle before the batch runs and measure the disagreement first — DK-9 (the differential-oracle harness) with DK-61's known-true case
65ruleR105DK-70a registry lists a member only after the gate spoke for it; a listed member whose site does not share is a named defect — DK-70 (the health suite asserts the work was done)
66ruleR106DK-71a policy taken on a remembered precedent is a belief — read the target project's tree — DK-71 (what earns belief)
67acceleratorP37 S106DK-65a stress-test before the fleet run and an audit's rejections read before believed: the struct-spelling byte finding corrects DK-65 (per-access spelling under the gate); the probe-first order and the known-true discipline are DK-61; DK-65
68acceleratorP37 S107DK-65the engine's first day: a linked-identity oracle is the build's own link (DK-61's known-true discipline, on a linker); a control that builds is a writer of build/ and the snapshot refuses what it cannot reproduce (DK-65's byte-gated spelling needs a baseline it can trust); process pools before threads, a selftest's scratch never live, typed signatures after the sites are members — engine-building lessons that ride with DK-65