title: "Briefing — "What should we investigate?"" source: "tasks/TFW-53__hl_contract_and_goal_defence/research/iter2/1_briefing.md"
Briefing — "What should we investigate?"¶
Mindset: Strategist. You're planning an investigation, not doing it. Frame what matters. Resist solving. Test: "Can I explain WHY we're investigating this and what would change our approach?" Parent: HL-TFW-53 — 🔒 FROZEN, baseline re-frozen after A1–A5 (
d9a4c57) Goal: An approved HL is a contract, and the reviewer is its defender — review asks "is this what we set out to do?" against a north star that sits above the task. Predecessor:research/iter1/RES.md— decisions D1–D14 constrain this iteration. Mode: Pipeline · deep (loops_per_stage: 3, counter-evidence required) · autonomous run, gates disabled by owner directive («tfw-53 автономно без вопросов deep mode iter 2»)
Predecessor Context (iteration 1)¶
Iteration 1 priced the freeze and redesigned its granularity. What it hands to this iteration:
| Constraint from iter1 | Effect on iteration 2 |
|---|---|
| D2/D4 — the frozen unit is the declarative claim, not the section text | The goal check measures against declarative claims, not against section prose. A wording delta is not a goal failure by construction |
D5/D6 — contract baseline = header + §12 + a reserved [agent/TFW-NN/freeze/coordinator] commit |
The goal check's reference set is retrievable: git log --grep='/freeze/' gives the reviewer the exact approved text |
| D10 — Phase HL becomes derivation-only (A1 approved) | The goal check must exclude the phase HL from its reference set. TFW-48 proves why: master P7 did not survive into the phase HL its reviewers checked against |
D11 — ❌ REJECT branch (a) redefined as file an amendment (A5 approved) |
Any new verdict class must compose with the amendment protocol, not around it |
D8/D9 — Type column, APPLIED — restrictive (A2 approved) |
Precedent that a verdict vocabulary carries consequences, and that the type must be visible at ruling time rather than reconstructed from prose |
| Open thread 3 — salami residual, unresolved | Carried here as (f3): does a git diff against the freeze baseline at the pre-TS gate cost anything? |
Also inherited: SS1 — the owner's interruption budget is spent on frequency, not on authority. A goal check that fires often is not a cheap check; the design's success metric is signal-per-fire.
Research Plan¶
Gather — "What do we NOT know?"¶
- Measure the review surface directly: how much purpose vocabulary exists in
review.md, the three stage templates and the three mode files today. Word-level, not impressionistic. - Resolve H12 mechanically: which file does PV Index priority 1 ("README Values") actually resolve to, in TFW and in a second real TFW project (AFD). Compare the two READMEs byte-for-byte where they overlap.
- Measure H13: the reviewer-relevant fraction of AFD's 509-line north-star anchor, by section and line count.
- Recover the negative-control corpus: the TFW-48/49 phase REVIEWs at
721ca15, their verdicts, and the master HL text they should have been measured against (9e19a4ffor TFW-49). - Decompose the problem into independent Dimensions — anchor locus, payload, obligation, forcing-function form, verdict class, materiality mechanism, reference set. Each must carry ≥3 alternatives with none marked preferred.
- External: standards-grade vocabulary for the "built the right thing" axis, and evidence on what makes a review gate degrade into a rubber stamp.
Extract — "What do we NOT see?"¶
- Build the Configuration Space across those dimensions; keep every combination that is not obviously contradictory.
- Cross-reference AFD's three concurrent
P{n}namespaces against the owner's own directive to test DoD-24's premise. - Model what each configuration costs a small project — the H12 adoption-tax question stated as a number, not a worry.
- Locate the retention failures: rules this framework already wrote and already lost (D46's "not rubber stamp", TFW-48's DoD-11/P7). If the pattern is retention rather than invention, the design target changes.
Challenge — "What do we NOT expect?"¶
- Pairwise consistency across the dimensions; eliminate the incompatible combinations and name why.
- The replay (HL DoD-26). Run the drafted check against six TFW-48/49 phase REVIEWs (seven verdicts) and against three TFW reviews that were genuinely sound. Required outcome: ≥1 non-approve on the former, 0 on the latter.
- The hardest negative control (f2): TFW-48's approved master already carried DoD-11 and P7 goal defence and lost both. Would the drafted check have fired given that the master already demanded it? A check that would not fire here is not a check.
- Attack the verdict vocabulary: does each candidate term survive translation out of software, and does it collide with a term the framework already uses?
- Counter-evidence duty (deep mode): actively look for the reading under which the north star is unnecessary, the forcing function is ceremony, and the check should not ship.
Hypotheses (from HL §10)¶
| # | Hypothesis | HL Status |
|---|---|---|
| H11 | The forcing function — the reviewer must quote the north-star clause the work serves — is what separates a live check from a rubber stamp. Without it the honest answer is "aligned" ~145 times out of 149 and the check decays within a few tasks | open |
| H12 | The project north star is the root README.md, and no new file is needed. Corollary: PV Index priority 1 "README Values" resolves to .tfw/README.md — methodology values identical in every project — not to project purpose. If so, priority 1 is misnamed in every TFW project |
open — owner's instinct, needs verification |
| H13 | A one-page north-star payload is sufficient for review purposes; AFD's 509-line anchor is ~41 reviewer-relevant lines, so the framework must not mandate the larger form | open |
Not re-litigated (settled upstream, per iterations.yaml): H2, H7, H8, H9, H10.
Resolved in iteration 1 and treated as constraints, not questions: H1, H3, H6.
Scope Intent¶
- In scope: the north-star anchor (locus, payload, obligation); the substantive goal check in Judge and its three
mandatory clauses; the verdict vocabulary for a goal failure and its finding; the reference set the check reads;
the
P{n}namespace guard; the DoD-26 replay validation; the salami residual (f3) as a cost question only. - Out of scope: anything in Phase A/B enforcement mechanics already decided by iteration 1; the AT execution mode (TFW-54, DoF-4); a TS→HL traceability gate (owner ruled out, DoF-6); Phase E trace restoration (needs no research); re-opening H8's "not a fifth stage" ruling — the owner's ruling on form is a dependency, not a research question.
- Explicitly not written by this role: HL, TS, code. Findings that touch frozen sections leave here as Amendment Proposals, per the mechanism this task is building and which its own HL header applies.
Guiding Questions¶
Owner directed an autonomous run. These are recorded, answered from evidence where possible, and routed to the coordinator in RES § Open Questions where they need a verdict.
- If PV priority 1 is misnamed in every TFW project (H12 corollary), is correcting it inside TFW-53's scope, or is it a separate defect that this task merely discovered?
- Must a project have a north star before it can run a review — or does the check degrade gracefully to the master HL §1 when no anchor is designated?
- Does a goal failure deserve a verdict name distinct from
❌ REJECT, given that the board must distinguish "built wrong" from "built the wrong thing" — and given that A5 already redefined REJECT's branch (a)?
User Direction¶
- Run autonomously, deep mode, no interactive gates (owner, 2026-08-08). All four 🛑 WAIT gates are executed as self-checkpoints and recorded in the stage files.
- Iteration 1's five amendment proposals were ruled on before this iteration started — all five APPROVED, applied, and re-frozen. This iteration therefore runs against an amended contract, and the amendment protocol has now been exercised once end-to-end on the task that invents it.
- Owner instinct on record (S26): «по идее сам ридми должен быть им» and «у нас тут есть два readme… кто из них является north-star?» — H12 exists because the owner asked this question, and it is the highest-value item here.
Stage complete: YES