title: "Briefing — "What should we investigate?"" source: "tasks/TFW-53__hl_contract_and_goal_defence/research/iter1/1_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 Goal: An approved HL becomes a frozen strategic contract that research may only amend through a logged, evidenced, owner-ruled channel.

Iteration: 1 of 2 (min) / 3 (max) · Mode: deep (loops_per_stage: 3, counter-evidence required) Run mode: autonomous, owner-authorised — /tfw-research tfw-53 автономно без вопросов deep mode. Stage WAIT gates are executed as self-checkpoints; no questions are asked mid-run. All questions are surfaced in RES for the coordinator.

Contract Notice¶

The parent HL is 🔒 FROZEN (approved 2026-08-08, baseline commit 8136306). Frozen: §1, §3, §4, §5, §6, §7. Free: §2, §7.2, §8, §9, §10, §11.

This research classifies its output accordingly. Findings that touch a frozen section are written as Amendment Proposals for HL §12, never as updates. This is the mechanism under investigation applied to itself — and it is a live test: if the protocol is unusable by the researcher who is designing it, that is a finding, not a formatting problem.

Research Plan¶

Gather — "What do we NOT know?" - Build the full corpus of historical HL Update Recommendations rows across every RES in tasks/ (the HL named six tasks; the corpus is larger and the wider count is strictly better evidence for H1). - Classify each row against the frozen/free split in HL §3 and produce a number: escalations per iteration. - Locate every Phase HL that has ever existed in this repository (working tree + git history) and diff it against its master HL (H6). - Enumerate the mechanism alternatives for contract state (H3) against D31, TFW-50's commit convention, and the git object model. - External: how mature change-control regimes bound an approved baseline — ADR immutability/supersession, CCB baseline management, repository-level ownership gates. - Decompose into independent Dimensions. Candidate factors already visible: freeze scope, freeze granularity, state mechanism, classification authority, escalation batching, freeze asymmetry, Phase HL governance, REJECT composition.

Extract — "What do we NOT see?" - Cross-reference the dimensions into a Configuration Space; look for the combination nobody proposed — specifically whether the frozen/free axis is the right axis at all, or whether change type and granularity carry more of the load than section identity. - Sub-classify the frozen-targeting corpus by change nature (deliverable precisification / scope addition / goal redefinition / acceptance change / principle change) and by direction (expansive vs restrictive), because the escalation cost of the design depends on that distribution, not on the raw count.

Challenge — "What do we NOT expect?" - Pairwise consistency across dimensions; eliminate incompatible combinations. - Counter-evidence, deliberately: the corpus was produced under D19, which ordered research to rewrite the HL. Would the same findings exist under a freeze? Selection effect must be priced in before the H1 number is trusted. - Stress the surviving configurations against DoF-2 (escalation spam), DoF-7 (undefined rejected path), DoF-10 (unverifiable baseline), and the documented CCB failure pair (rubber stamp / bottleneck). - Resolve freeze asymmetry and REJECT composition, the two open design questions in HL §10 Blind Spots.

Hypotheses (from HL §10)¶

#	Hypothesis	HL Status
H1	In TFW's own history, the large majority of RES `HL Update Recommendations` targeted free sections (§2, §8, §9, §10) rather than frozen ones — so freezing six sections costs few escalations per task	open — one counter-signal (TFW-49 `642c647`); needs the full corpus count
H3	A contract state field in the HL header plus an append-only §12 is sufficient state; no filesystem-level marker (lock file, approved-HL snapshot) is needed despite D31	open
H6	Phase HLs in multi-phase tasks are a real drift channel: historical Phase HLs introduced deliverables absent from their master HL	open — TFW-48/49 phase HLs believed to show no content drift

Also in scope from HL §10 Blind Spots (not numbered hypotheses): - Freeze asymmetry — do tightening a DoF and loosening a DoD need the same amendment path? - REJECT composition — how the amendment protocol composes with the existing ❌ REJECT → user branching point in conventions.md §5.

Settled upstream and not re-litigated: H2 (confirmed), H7 (confirmed), H8/H9/H10 (settled by AFD recon). H11/H12/H13 belong to iteration 2.

Scope Intent¶

In scope: the contract half of TFW-53 — freeze scope, freeze granularity, contract-state mechanism, classification and escalation protocol, Phase HL governance, amendment/REJECT composition. Evidence is this repository's own history.
Out of scope: goal defence in review (north star, Judge check, reviewer identity, verdict vocabulary, replay validation) — that is iteration 2 in full. Phase E trace restoration needs no research. AT delegation mode is TFW-54.
Method boundary: this iteration proposes; it never edits the HL, and it never writes TS.

Guiding Questions¶

Autonomous run — no questions asked of the user during this iteration. The three questions this iteration would otherwise have put to the owner are carried into RES.md § Open Questions for the coordinator to route:

If freezing all six sections escalates on ~every iteration, is the correct response to shrink the frozen set or to shrink the frozen unit inside §3/§4?
Should tightening a constraint (adding a DoF, narrowing scope, dropping a deliverable) travel the same amendment path as loosening one?
Do Phase HLs inherit the freeze, or should the artifact class stop existing?

User Direction¶

Owner instruction for this run, recorded verbatim: «tfw-53 автономно без вопросов deep mode» — run iteration 1 autonomously, in deep mode, without interactive gates. Owner also fixed the run architecture in HL §11 S27: research runs in a separate session with no planning-conversation history, deliberately, because in TFW-49 the same coordinator that ran the research accepted it.

Stage complete: YES