| name | interactive-results |
| description | Turn a completed bakery or restaurant rubric run into a validated local faceted list-and-map website; use only after Phase 8 and publish only on a separate explicit request. |
Interactive Results
Invoke as /interactive-results <RUN_DIR>. The path is mandatory: if it is missing, ask the user for the exact run directory and stop. Never guess, infer, auto-select, or create a run directory.
This workflow is snapshot-only. It transforms an already completed bakery or restaurant run; it must not perform fresh web lookup, research, scoring, rating, access, location, recommendation, or evidence work. Never change an orchestrator decision. If the snapshot cannot support a field, leave it null, omit it with a recorded reason, or ask about a consequential ambiguity as defined below.
Hard-gated lifecycle
- Resolve the user-supplied path and keep every durable output inside it.
- Read
00-run-manifest.md. Require Phase 8 completion (complete) and required artifacts 05-evidence-ledger.md, canonical 06-decisions.json, and 08-results.md. Reject an incomplete or category-unknown run.
- Validate
06-decisions.json against decision-schema.json, its source references, IDs, merge targets, and manifest metadata. It is the only canonical decision ledger.
- For a legacy completed run without JSON, perform the one-time migration below. Never treat independently authored
06-decisions.md as a second authority.
- Build a snapshot projection accounting for every canonical ID as Practical, Audit, or omitted with a reason. Prefer descriptions already written in
08-results.md; otherwise make a faithful concise summary only from canonical rationale and accepted evidence.
- Present the required facet-design checkpoint and stop for approval, revision, or the exact instruction
use your defaults.
- Write the approved
interactive-results-projection.json, validate it against interactive-schema.json, then run the deterministic generator. Do not rewrite template.html or application JavaScript per run.
- Serve locally and complete static plus browser validation at desktop and mobile sizes.
- Write
interactive-results-validation.json, then update the manifest with artifact paths, schema versions, hashes, generation timestamp, browser results, and final status. A failure cannot be recorded as complete.
- Return local paths and a validation summary. Publication is a separate permission boundary.
Legacy migration
If 06-decisions.json is absent but the manifest proves a completed legacy run, transform 06-decisions.md, 05-evidence-ledger.md, 08-results.md, and the manifest once. This is an agent-guided migration, not a claim that arbitrary Markdown has a deterministic parser.
First write {RUN_DIR}/interactive-results-migration.json containing parsed rows, source references, unresolved fields, duplicate or missing IDs, and a consequential_ambiguities list. Stop and ask only when ambiguity affects identity, disposition, Practical eligibility, score/rating display, or accepted map location. Do not research the answer. After resolution, write canonical JSON atomically only when schema and cross-record invariant validation pass, then record the migration report and canonical hash in the manifest.
Practical, Audit, and location rules
- Practical contains all currently obtainable qualifying records in the canonical snapshot, including stable walk-in, preorder, market, delivery, or other accepted acquisition paths. It is not merely a top-three display.
- Audit contains non-Practical canonical records useful for transparency: filtered, disqualified, unavailable, unresolved, unconfirmed, or duplicate dispositions. Use compact rows. Use
Directions only for Practical records with an accepted address. Otherwise, Find on Google Maps may search by canonical name plus run scope; label it as a search, never as an accepted location.
- Omitted is allowed only with a per-ID reason; populations must not overlap.
- Use only accepted business addresses and coordinates from
06-decisions.json. Never geocode, repair, or guess coordinates. A no-coordinate corpus must render as list-only.
- Ratings, scores, access, names, and dispositions must match the canonical snapshot exactly.
Required facet-design checkpoint
Before generation, show a proposed facet table with each facet, controlled values, representative venues, aliases/normalization, local extensions, and rejected sparse or overlapping values. Cover shared region, access, occasion, and status dimensions, plus evidence-supported category dimensions such as specialty for bakeries or cuisine/format for restaurants. Do not invent unsupported labels.
Ask the user to approve, revise, or reply use your defaults, and stop. Approval is required even when the defaults appear obvious. After approval, account explicitly for every canonical ID and freeze the projection; no fresh research follows.
The interactive occasion facets represent the full evidence-supported suitability set, not only the three venues printed under each occasion in 08-results.md. Preserve those top-three recommendations as highlights, then expand the facet memberships from accepted decision fields and record the deterministic rules and before/after counts in {RUN_DIR}/interactive-occasion-assignment-audit.json. This derived expansion must not change 06-decisions.json, the Markdown shortlist, scores, or rankings. Apply the same split to rare finds: broaden only from explicit accepted item/scarcity evidence, never from cuisine, specialty, or a high score alone.
Deterministic generation
From the repository root, run:
python3 interactive-results/generate.py \
--decisions "$RUN_DIR/06-decisions.json" \
--projection "$RUN_DIR/interactive-results-projection.json" \
--template interactive-results/template.html \
--output "$RUN_DIR/index.html" \
--validation-output "$RUN_DIR/interactive-results-validation.json" \
--generated-at '<ISO-8601 timestamp>'
The initial report is a static validation result. Regenerate with the same inputs/timestamp and verify an identical output hash. Syntax-check the inline application script with Node and parse the HTML with Python's html.parser.
Local browser validation
Serve the generated file over HTTP, not file://:
python3 -m http.server 8765 --directory "$RUN_DIR"
Follow interactive-results/tests/browser-checklist.md. Browser validation is mandatory at 1440×900 and 390×844 and must test real search, facets, alternate counts, rating floor, sorts, Practical/Audit modes, localStorage shortlist persistence, map/list navigation, drawers, keyboard close/focus behavior, independent scrolling, and list-only fallback. Capture page errors and screenshots. HTTP 200 alone is insufficient.
Merge browser outcomes into interactive-results-validation.json, set final status only after every required assertion passes, and include canonical decision, projection, template, and output SHA-256 hashes. Update 00-run-manifest.md only afterward.
Publication boundary
Default delivery is local index.html; there is no automatic publication, gist creation, upload, or network sharing. At completion, you may offer:
/gisthost "$RUN_DIR/index.html"
Do not invoke /gisthost unless the user makes a new, separate explicit request.
