| name | haipipe-application-claims |
| description | Stage orchestrator for rung 1c of the venue-FREE evidence ladder (1a-descriptions -> 1b-themes -> 1c-claims -> 1d-advice): the claim ledger + evidence campaign brain — what generalizes, with status + anchor. THE ONLY home of a claim's status (supported | weak | GAP). Three sections: Claims (theme-tagged, status + PP reference), Q-consumer (the evidence questions raised, one `## Q-Claim-<n>` block each), Evidence Campaign (dispatch order + dependencies). The pinned venue sets how much of the campaign must SETTLE (light/medium/full). Markdown only. Trigger: claims, claim ledger, what must be true, what generalizes, evidence plan, probes, supported, GAP, /haipipe-application claims. |
| argument-hint | [intervention-path] [--backfill <PPNN>] |
| allowed-tools | Bash, Read, Write, Edit, Grep, Glob, Skill |
| metadata | {"version":"0.7.6","last_updated":"2026-07-19","summary":"Claims stage (rung 1c of the venue-FREE 1a–1d evidence ladder; the K rung) — the ONLY home of a claim's status (supported | weak | GAP, from a judged artifact, never intuition; Consumes 1b theme hooks into a claim ledger + evidence campaign, gives each primary a Rival + refute-capable probe, and reaches for evidence via the PROBE phase (questions raised as entries in 1-probes/). The pinned venue sets how much of the campaign must SETTLE (light/medium/full). History: ./CHANGELOG.md."} |
Skill: haipipe-application-claims
Rung 1c of the venue-FREE evidence ladder, and the rung that reaches hardest for evidence.
It answers: which claims generalize — supported, weak, or GAP — and what evidence settles each?
1a-descriptions what the data looks like
1b-themes what patterns/topics emerge
1c-claims what generalizes (the ledger) <- THIS RUNG
1d-advice what the evidence advises (the deliverable)
Read first: ../../../PHILOSOPHY.md, ../../haipipe-application-lifecycle/SKILL.md (Intervention Lifecycle Contract), ../../../haipipe-application/SKILL.md (Delivery Need Routing).
What's special: three things make claims claims
1. It is the ONLY home of a claim's status.
supported | weak | GAP lives HERE, per-claim, private to this intervention — never in a probe file.
A probe entry carries the evidence (its ### a-executor, a copy of the answering QA file's answer); the judgment is ours to write.
A claim is supported only when it traces to a JUDGED artifact (a probe answer, or an equivalently reviewed result) — never from intuition, never from a raw unjudged number.
2. The venue sets the settlement bar, read at the gate.
Venue-FREE: retargeting changes the required settlement, never the ledger's truth.
CHECK reads STATUS.md | claims_settlement | and evaluates the Evidence Campaign table against it (light | medium | full — table below).
The same bar reaches 1d: an advice entry may only cite claims at/above it.
3. It consumes 1b's theme hooks, and it tests its primaries to destruction.
Every 1b theme hook becomes a C<n> entry or a Declined-hooks line with a why — the Declined section is the reservoir the next round re-mines.
Every primary claim carries a Rival: (the strongest alternative explanation) and a refute-capable probe (Refutes-if: — the result that would FLIP it, not only confirm it).
There is NO Hypotheses section (application delta vs paper): the mechanism lives in seed/pitch, the theme space in 1b, and alignment is in the tags (C1 (T1), PP03 (C1/C3)).
The four phases, in claims
DRAFT re-mine last round's Declined reservoir, read 1b-themes.md (hooks = claim candidates)
and 1a-descriptions.md (the grounding floor); CONSUME every hook; give every primary a
Rival; write Claims (short, theme-tagged), Q-consumer (the evidence questions), Evidence Campaign
PROBE one worker call; the five-step loop raises each GAP/weak claim as a SECTION in 1-probes/
and COLLECTS — the claim status FLIPS at INTERPRET, numbers land INLINE on the claim (+ resolving pointer).
Routing mechanics are the probe layer's: ../../../2-phase/1-probe/haipipe-application-probe/SKILL.md
REVISE refine claim statements, probe-plan clarity, campaign ordering
CHECK settlement bar vs the campaign table; every hook consumed; every primary has a Rival +
Refutes-if; every settled claim traced to a RESOLVING QA file; no aspirational anchors; no STALE tags
Claims RECEIVES evidence, never PRODUCES it inline (LAW 1): it raises questions; haipipe-application-probe binds them.
--backfill <PPNN> flips a claim from a landed answer: supported -> supported · refuted -> drop or reword · inconclusive -> stays weak/GAP with the caveat recorded.
Rounds + back-routing (loop-until-dry for medium+ venues, [ROUND n] / [ROUTE -> themes|descriptions] in _LOG) follow ../../../haipipe-application/SKILL.md (Stage Gate Protocol).
Seed's [FORWARD -> CLAIMS] pointers are consumed by rung 1a, which seeds this doc's questions; this rung no longer greps seed's _LOG itself.
The artifact
0-lifecycle/1c-claims/1c-claims.md — full skeleton in ref/claims-template.md:
Claims one C<n>: statement + theme tag (T<n>) + role (primary|enabling|assumption)
+ status + -> PP reference; primaries carry a Rival line (short; the question is in Q-consumer)
Declined hooks theme hooks considered and not committed, one line + why (the reservoir)
Q-consumer one `## Q-Claim-<n>` block (Ask / Why / Answer=__TO_BE_FILLED__) per evidence question this stage raises; the route,
approver, and Refutes-if organize into 1-probes/ at APPROVE, not here
Evidence Campaign dispatch order + a compact summary table (the gate reads this)
Sidecar: _LOG_1c-claims.md (phase journal). Verified numbers stay INLINE on the claim (+ a resolving pointer to the task/discovery result, which is the source of truth) — no _VALUES_ / _CITATION_ sidecar.
A claim with no plausible theme parent signals a 1b gap: loop back to themes rather than orphan-tagging.
Settlement Gate (venue-scaled, read at CHECK against the campaign)
light (sms, push, reminder) every claim the artifact leans on is at least tied to a named
anchor or "common knowledge"; GAPs allowed if not load-bearing
medium (checklist, email) primary claims supported or weak-with-caveat;
load-bearing GAPs have campaign rows (dispatch optional)
full (dashboard, ui-card, primary claims supported (judged); every GAP in the campaign,
report) load-bearing ones settled
Absent claims_settlement (venue unpinned) → apply light provisionally.
Questions this stage typically raises
DRAFT's RAISE+PLAN step raises what the draft cannot answer. These are the kinds this stage is prone to — read this list, then walk the draft against it.
⚖️ sufficiency Does this evidence support the claim, or only fail to
contradict it? Name the effect and the N.
🔀 rival explanation What else would produce this pattern, and what would
distinguish them? A claim with no named rival is untested.
🎯 generalization Does it hold for the audience we will actually deploy to,
or only for the sample it was found in?
🧱 ingredient Is a prerequisite still missing? Then the claim is
BLOCKED-ON-DESCRIPTIONS — do not re-ask 1a's question here.
Exits
promote -> /haipipe-application advice derive the advice (the ladder's deliverable)
End every reply with the closing block (stage line via ../../../haipipe-application/stage-strip.sh).