title: "Briefing — "What should we investigate?"" source: "tasks/TFW-60__conflict_resistant_shared_workspace/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-60 Goal: Independently attack the iteration-1 Phase-A architecture so concurrent humans and agents can rely on task-local state, coordination, file sync and Git landing without losing discoverability, provenance or stable traces. Predecessor: Iteration 1 RES and stage traces

Research Plan¶

Gather — three adversarial evidence loops¶

T1, corpus and discovery: build a deterministic 100-task mixed valid/legacy/malformed fixture outside tracked project paths; exercise normal, absent, stale and malformed derived indexes; record a fresh-agent cold-start path separately from heuristic reasoning and from genuinely observed non-technical-human evidence.
T1, schema and readers: challenge every proposed status.yaml field against a named reader and lifecycle trigger; inspect current primary comparable-system sources for evidence that central or richer state, Markdown, markers or databases solve a failure C1-R cannot, without reopening eliminated families by preference.
T2, journal and recovery: gather deterministic fixtures for owner_epoch, ownership recovery, event-first/snapshot-second ordering, identical/divergent duplicate IDs, malformed records, segment rollover, terminal rejected traces and coordinator disappearance; measure the provisional 240-code-point / 100-event / 32-KiB bounds without promoting them to defaults prematurely.
T3, Git and migration: inventory the repository's Full/Assisted legacy and nonstandard task shapes read-only, including proposal-without-HL and malformed board rows; create temporary Git repositories/worktrees for G-B and guarded G-A attacks covering metadata preflight, unrelated peer changes, staged allowlists, single-task commits and producer/landing-owner attribution.
Evidence boundary: distinguish repository/runtime observations, deterministic temporary fixtures, primary-source contracts, usability heuristics, real human observations and actual provider runtime observations. Never let a local fixture claim vendor sync behaviour or a researcher proxy claim human usability.

Extract — cross-reference what survives¶

Cross-reference T1 dimensions: catalogue state (normal/absent/stale/malformed), task shape (valid/legacy/malformed), reader (fresh agent/file-browser human/workflow), authority decision and status-field reader/trigger coverage.
Cross-reference T2 dimensions: owner epoch, recovery authority, journal/snapshot visibility order, event-ID condition, malformed record, rollover threshold, terminal outcome and coordinator availability; derive deterministic stop/recovery outcomes without timestamp authority.
Cross-reference T3 dimensions: G-B/G-A profile, Git-dir/worktree/index preflight state, peer-change timing, staged allowlist result, task/phase scope, producing actor, landing owner and journal/commit ordering.
Separate the shared semantic lifecycle/ownership/event contract from Full- and Assisted-specific artifact, transport and Git profiles; reject any edition split that changes shared meanings without counter-evidence.
Map every legacy/nonstandard task shape to verified facts, explicit unknowns and compatibility behaviour; do not invent missing HL/status facts or propose path normalization that moves existing tasks.

Challenge — seek disconfirmation before synthesis¶

Attack C1-R with the 100-task mixed corpus across normal, absent, stale and malformed index cases; test whether fresh-agent selection is reproducible and whether the zero-command file-browser route remains merely a heuristic when no real non-technical participant is available.
Attack journal recovery with all named failure fixtures, multi-byte Unicode and multiple rollovers; compare the provisional bounds by measured encoded sizes/reader costs and preserve real-provider reconnect as an acceptance gap unless actual runtime evidence exists.
Attack G-B and guarded G-A in temporary repositories/worktrees: wrong/missing GIT_DIR, unexpected worktree/index, unrelated peer changes, exact staged-path allowlists, staged-versus-working changes, one-task commit scope and producer/landing-owner provenance.
Resolve landed ordering by comparing a post-commit event, commit-only provenance and a pre-landing handoff/reference design against resume, release, crash recovery, dirty-tree and circular-reference failures.
Challenge lossless migration and shared-edition semantics against the full read-only legacy census. Reopen C2–C5 or G-C only if new counter-evidence defeats a C1-R/G-A/G-B invariant rather than because another carrier appears more familiar.

Predecessor Decisions Under Attack¶

Decision	Iteration-1 result to validate adversarially
D1	H1 was refuted as stated; retain a permanent router plus persisted, rebuildable, derived `tasks/INDEX.md` for zero-command portfolio discovery.
D2	C1-R was the sole surviving Phase-A control configuration: strict `status.yaml`, separate numbered journal segments, derived index, stable paths, one state owner and explicit recovery.
D3	Authority, projection, journal and Git are distinct responsibility layers; file sync is transport and grants none of their authority.
D4	Status is a small validator-owned YAML subset with no duplicate keys, anchors, aliases, custom tags, free-form notes or copied artifact bodies; `owner_epoch` detects stale authority.
D5	The shared lifecycle kernel is `new`, `active`, `waiting`, `blocked`, `terminal`, with `done`/`rejected` terminal outcomes and edition-specific `workflow_stage`.
D6	Full Coordinator or Assisted steward is the sole normal state/journal writer; recovery increments `owner_epoch` through a predeclared authority.
D7	Journal grammar is `created`, `dispatch`, `handoff`, `transition`, `ownership_changed`, `amendment_escalated`, `landed`, plus reserved `consolidation`; block/resume are typed transitions.
D8	Events carry monotonic ID, UTC time, kind, actor, epoch, applicable delta, references, optional related event and one bounded optional summary; corrections append.
D9	Transitions are event-first/snapshot-second; journal-ahead may be recoverable, snapshot-ahead is invalid, and identical duplicate IDs alone are idempotent.
D10	History is segmented and retained, but 240 code points, 100 events and 32 KiB are hypotheses rather than established constants.
D11	Normative file-sync rules assume independent-file propagation only; no cross-file ordering, lock, transaction or provider-specific guarantee.
D12	Migration is compatibility-first: preserve legacy paths and verified facts, report malformed/nonstandard tasks and never normalize by moving active/history-bearing directories.
D13	G-B is the baseline one-landing-owner Git profile; guarded G-A is optional Full; Git administration never synchronizes.
D14	Landing uses literal task paths, an exact staged-name allowlist, a staged-versus-working recheck, one task/phase scope and separate catalogue landing.
D15	C2–C5 and G-C remain eliminated unless new evidence defeats the surviving invariants.
D16	Iteration 1 found no frozen-HL amendment necessary; all recommended mechanisms fit the approved Phase-A contract.

Open Threads From Iteration 1¶

Thread	Required closure or explicit gap
T1 — hybrid catalogue and strict YAML	Test a 100-task mixed corpus and four index conditions; supply fresh-agent evidence; distinguish usability heuristics from real non-technical-human observation; challenge every status field with a reader and trigger.
T2 — journal and recovery	Test epoch/recovery/order/duplicate/malformed/rollover/rejection/disappearance; evaluate numerical bounds reproducibly; keep real provider behaviour separate from deterministic simulation.
T3 — Git landing and migration	Test G-B and guarded G-A failure/scope/provenance cases; settle `landed` ordering; census Full/Assisted legacy and malformed shapes read-only; preserve shared semantics across edition profiles.

Iteration-1 open questions Q1–Q6 map directly to these threads: human/fresh-agent usability (Q1), numerical limits (Q2), offline/provider recovery evidence (Q3), landed ordering (Q4), guarded G-A exposure and failure contract (Q5), and exhaustive compatibility migration (Q6).

Hypotheses (from HL §10)¶

#	Hypothesis	HL Status Entering Iteration 2
H1	A persistent Task Board is not required: a standard on-demand command can assemble task ID, goal, value, live status and terminal outcome from task-local sources without degrading cold-start agent planning or human discovery.	Iteration 1 refuted it as stated and confirmed a hybrid; iteration 2 attacks the permanent-router + persisted-derived-index replacement.
H2	A tiny machine-readable task-local status carrier — marker files or a strictly bounded YAML schema — is safer than a mutable Markdown status page and remains understandable to non-technical users through normal file browsing.	Iteration 1 partially confirmed strict reduced YAML and eliminated markers/Markdown; direct non-technical editing remains unobserved.
H3	A separate coordinator-owned append-only journal with a closed event vocabulary, artifact references and a size/retention rule preserves cross-session management context without duplicating canonical artifacts or becoming a new writing surface.	Iteration 1 supported it subject to recovery and bounds evidence; iteration 2 attacks both.
H4	Assisted and Full can share one task-local state/journal contract while differing only in collaboration transport and Git requirements; separate task models are unnecessary.	Iteration 1 confirmed shared semantics while allowing edition-specific artifact, transport and Git profiles; iteration 2 audits legacy counter-evidence.

Scope Intent¶

In scope: TFW-60 Phase A iteration 2 only; adversarial validation of C1-R, the hybrid catalogue, strict status schema, coordinator journal and recovery, G-B/guarded G-A Git landing, lossless Full/Assisted migration, and the shared semantic contract; current primary technical sources; deterministic temporary fixtures outside tracked project paths; read-only repository/legacy census; H1–H4 verdicts and classified HL recommendations.
Out of scope: Phase B debt architecture; Phase C knowledge architecture; implementation or TS design; edits to the frozen HL, research/iterations.yaml, README, framework, adapters, legacy tasks or any artifact outside research/iter2/; iteration 3; destructive operations in external synchronized folders; claiming a deterministic simulation as provider runtime evidence; claiming a researcher simulation as observed non-technical-human evidence.

Guiding Questions¶

What new evidence, if any, defeats the C1-R authority/projection/journal separation or the permanent-router + persisted-index hybrid strongly enough to reopen an eliminated configuration family?
Which journal recovery rules and bounds are reproducibly supported, and which must remain Phase-A TS acceptance gaps because real provider or human evidence is unavailable?
Which Git landing/provenance order and compatibility rules survive G-B/G-A fixtures and the complete read-only legacy census without splitting the shared semantic contract by edition?

User Direction¶

Owner authorized the full autonomous research cycle through delegated Codex tasks; the parent task is the Coordinator. The Researcher reports and stops at every TFW gate rather than asking the owner directly.
Coordinator approved deep mode for iteration 2: up to three OODA loops per stage, technical claims cross-checked and explicit counter-evidence required.
Iteration 2 is independent adversarial validation of iteration 1. C2–C5 and G-C may reopen only if new counter-evidence defeats C1-R, G-B or guarded G-A.
Required threads are T1 hybrid catalogue/strict YAML, T2 journal/recovery, and T3 Git landing/migration. Phase A only; debt and knowledge architecture remain excluded.
Only research/iter2/ stage traces and research/iter2/RES.md may be written. Human/provider evidence must be labelled honestly; unavailable real evidence remains a TS acceptance gap.

Stage complete: YES