{ "$schema": "https://json-schema.org/draft/2020-12/schema", "$id": "https://aaronzhang.ai/tie/schema/tie-daily.schema.json", "title": "TIE Daily Record", "description": "One day's published output of the Trend Intelligence Engine. This is a PUBLIC PROJECTION of the internal DigestReport, not a copy of it: internal routing, delivery and audience-lens fields are dropped at export time, never redacted after the fact.", "type": "object", "additionalProperties": false, "required": [ "schema_version", "date", "run_id", "executed_at", "domain", "locale", "window", "generated_at", "status", "trends", "coverage", "provenance", "license" ], "properties": { "schema_version": { "const": "1.0" }, "date": { "type": "string", "format": "date", "description": "The day this record is ABOUT: the END of the observation window, as a calendar date in America/Los_Angeles. Also the filename stem, and the date in the human-readable header. Use this for anything reader-facing. Note it will NOT match the date you get by slicing `window.to`, which carries an Australia/Sydney offset and is usually one day later — same instant, different calendar." }, "executed_at": { "type": "string", "format": "date-time", "description": "When the pipeline actually ran. Normally the same calendar day as `date`, but not guaranteed — a catch-up run can produce a record days later." }, "run_id": { "type": "string", "pattern": "^\\d{4}-\\d{2}-\\d{2}-ai_industry$", "description": "Opaque internal run identifier, carried through for traceability. ⚠️ THE DATE INSIDE run_id IS THE WINDOW START, NOT `date` AND NOT THE RUN DATE. For a 3-day window, run_id `2026-08-03-ai_industry` has date `2026-08-06` and executed on 2026-08-06. Never parse a date out of run_id — read `date`, `window`, or `executed_at`." }, "domain": { "const": "ai_industry" }, "locale": { "enum": ["en", "zh"], "description": "Language of the human-readable prose fields (trend_name, short_description, attention_driver, so_what)." }, "window": { "type": "object", "additionalProperties": false, "required": ["from", "to", "timezone", "days"], "description": "The observation window. AI runs use a 3-day rolling window: lab blogs and arXiv do not publish daily, so a 24h window is mostly empty and over-weights news aggregators.", "properties": { "from": { "type": "string", "format": "date-time" }, "to": { "type": "string", "format": "date-time" }, "timezone": { "type": "string", "examples": ["Australia/Sydney"] }, "days": { "type": "integer", "minimum": 1 } } }, "generated_at": { "type": "string", "format": "date-time" }, "status": { "enum": ["ok", "empty", "failed"], "description": "ok = at least one trend published. empty = the pipeline ran but synthesis produced nothing usable (output-side gate withheld it). failed = the run did not complete, most often because the input-side gate refused to emit on zero raw input. `empty` and `failed` days are PUBLISHED, with trends: []. Silently skipping bad days would make the archive look healthier than the pipeline is — and a `failed` day is usually the safety design working, not the engine guessing." }, "status_detail": { "type": ["string", "null"], "description": "Present when status != ok. Plain-language reason, e.g. 'zero raw input; cannot distinguish quiet window from unavailable sources'." }, "trends": { "type": "array", "maxItems": 3, "items": { "$ref": "#/$defs/trend" }, "description": "The day's headline trends. Capped at 3 by design — this is a reading aid, not a feed." }, "watch_list": { "type": "array", "maxItems": 5, "items": { "$ref": "#/$defs/trend" }, "description": "Detected but below the bar for `trends`. Published so that promotions into `trends` on later days are auditable." }, "trend_evolution": { "type": "array", "items": { "type": "object", "additionalProperties": false, "required": ["trend_name", "canonical_id", "prior_stage", "current_stage", "delta_description"], "properties": { "trend_name": { "type": "string" }, "canonical_id": { "type": "string" }, "prior_stage": { "$ref": "#/$defs/lifecycle" }, "current_stage": { "$ref": "#/$defs/lifecycle" }, "delta_description": { "type": "string" } } }, "description": "Lifecycle transitions since the previous run, keyed by canonical_id. This is what makes the archive a time series rather than 90 unrelated snapshots." }, "coverage": { "type": "object", "additionalProperties": false, "required": ["sources_configured", "source_keys_reporting", "signals_raw", "signals_accepted", "clusters_total"], "description": "Collection accounting for this run. Published so a reader can tell a quiet day from a broken one.", "properties": { "sources_configured": { "type": "integer" }, "source_keys_reporting": { "type": "integer", "description": "Distinct publishers that contributed at least one accepted signal (the keys of `by_source`). ⚠️ Not a count of registry entries: a company's feeds fold into one publisher, and items that arrive through an aggregator (AIHOT) count under their original publisher, so this can be smaller or larger than `sources_configured`." }, "signals_raw": { "type": "integer", "description": "Items fetched, before dedup." }, "signals_accepted": { "type": "integer", "description": "Items surviving dedup and scoring." }, "clusters_total": { "type": ["integer", "null"], "description": "Null when this run recorded no cluster accounting. That covers two different situations, and `status` does not distinguish them: synthesis never succeeded (`--no-llm`, an LLM failure, a budget degrade), OR synthesis succeeded normally but ran without the input-budget selection that produces these counts. A healthy `status: ok` record can carry null here." }, "clusters_synthesized": { "type": ["integer", "null"], "description": "Clusters that fit the synthesis context budget. clusters_total - clusters_synthesized were NOT seen by the model." }, "by_source": { "type": "object", "additionalProperties": { "type": "integer" }, "description": "Accepted signal count per publisher id. A company's feeds fold into one key (google, meta, openai); items that arrive through an aggregator (AIHOT) count under their original publisher (its mapped id, or a slug of its name). Not the same keys as sources.json." } } }, "provenance": { "type": "object", "additionalProperties": false, "required": ["pipeline_version", "config_hash", "prompt_hash"], "description": "Reproducibility surface, and deliberately nothing more. The pipeline source is not public; these three let a reader verify that two days came from the same configuration and the same prompts, and date the day a methodology changed. The model name, the reasoning effort and the per-run cost are not published: nothing here promises them, no reader can act on them, and an MIT-licensed file cannot be unpublished once it is forked.", "properties": { "pipeline_version": { "type": "string", "examples": ["tie-0.2.0"] }, "config_hash": { "type": "string", "description": "Hash over the source registry + runtime config. A change here means the input set or scoring parameters moved." }, "prompt_hash": { "type": "string", "description": "Hash over the synthesis prompts. A change here means the analysis instructions moved." } } }, "source_flags": { "type": "array", "items": { "type": "string" }, "description": "Machine-readable warnings raised during the run, e.g. 'dead_source:meta_ai_blog', 'llm_call_failed'. Published verbatim. This is the machine half of the run's flags: every entry starts with one of a known set of prefixes. The prose half is in `caveats`. Before 2026-08-10 both halves were published together under a single `risk_flags` key." }, "caveats": { "type": "array", "items": { "type": "string" }, "description": "The prose half of the run's flags — reader-facing sentences the synthesis step wrote about its own output, rather than machine warnings (those are in `source_flags`). Expected to be EMPTY on runs from 2026-08-10 onward: the channel that let the synthesis model author whole flags was closed at the parse boundary on that date. A non-empty array on a recent record means prose reached the flag channel by some other route." }, "style_flags": { "type": "array", "items": { "type": "string" }, "description": "What the style probe could not fix by itself, plus any notes from the English translation step. See METHODOLOGY §10 for what this does and does not mean — in particular, an empty array is not a claim that the prose is clean. Auto-fixed rewrites are not listed here." }, "insights": { "type": "array", "description": "Cross-trend observations: the issue's executive summary. Built by the render layer, so it reaches this record via the published markdown rather than via the internal digest.", "items": { "type": "object", "additionalProperties": false, "required": ["text", "refs"], "properties": { "text": { "type": "string" }, "refs": { "type": "array", "items": { "type": "integer" }, "description": "1-based positions in `trends` that this insight points at. Insights carry no inline links by design; these pointers are how an insight is tied back to its evidence." } } } }, "since_last": { "type": "array", "items": { "type": "string" }, "description": "The continuity ledger: which of the prior issue's headline trends advanced, held, or dropped out. Prose, one sentence per entry. `trend_evolution` is the structured view of the same idea." }, "countdowns": { "type": "array", "description": "Dated deadlines extracted from this window's signals — AI pricing changes, deprecations and policy enforcement dates. Re-rendered on every run until the date passes, so the same countdown appears on consecutive days with a decreasing `days_left`.", "items": { "type": "object", "additionalProperties": false, "required": ["title", "kind", "effective_date", "days_left", "description", "url"], "properties": { "title": { "type": "string" }, "kind": { "type": "string", "description": "Human-readable label for the deadline type (pricing change / deprecation / policy). Localized, not a stable enum — do not key on it." }, "effective_date": { "type": "string", "description": "ISO `YYYY-MM-DD` when stated by the source. NOT `format: date`: the empty string is emitted when no date could be recovered from the rendered line." }, "days_left": { "type": ["integer", "null"], "description": "Days from the issue date to `effective_date`, as counted at render time. Null when it could not be recovered." }, "description": { "type": "string" }, "url": { "type": "string", "description": "Source link, or the empty string when the rendered line carried none. NOT `format: uri` for that reason." } } } }, "leaderboard": { "type": "array", "description": "Benchmark position changes claimed inside this window, previous → new.", "items": { "type": "object", "additionalProperties": false, "required": ["benchmark", "from", "to", "url"], "properties": { "benchmark": { "type": "string" }, "from": { "type": "string", "description": "Prior holder and score, as a single string. Empty when the rendered line stated no prior." }, "to": { "type": "string", "description": "New holder and score. Scores are copied verbatim from the source, units included ('76.2%', '1410 Elo') — never normalized." }, "url": { "type": "string", "description": "Source link, or the empty string. NOT `format: uri` for that reason." } } } }, "sections": { "type": "object", "additionalProperties": false, "description": "The desk sections. Each is topic-triggered: a desk appears only on days its topic cleared a threshold, so a missing key means 'nothing qualified today', not 'this desk was removed'. `sections_missing` reports which headings were not found at all.", "properties": { "pricing": { "$ref": "#/$defs/desk" }, "media": { "$ref": "#/$defs/desk" }, "incidents": { "$ref": "#/$defs/desk" }, "tools": { "$ref": "#/$defs/desk" }, "frameworks": { "$ref": "#/$defs/desk" } } }, "sections_missing": { "type": "array", "items": { "type": "string" }, "description": "Render-layer modules whose heading was not found when this record was assembled. Normally a subset of `insights`, `since_last`, `countdowns`, `leaderboard`, `pricing`, `media`, `incidents`, `tools`, `frameworks` — low-frequency desks are expected here on a given day, whereas a missing `insights` or `since_last` is worth looking at. ⚠️ Deliberately NOT an enum: when the issue's markdown is absent entirely this carries the single sentinel `\"-\"`, which is not a module name." }, "license": { "const": "MIT" }, "author": { "type": "object", "additionalProperties": false, "required": ["name"], "properties": { "name": { "type": "string" }, "url": { "type": "string", "format": "uri" } } } }, "$defs": { "lifecycle": { "enum": ["emerging", "rising", "peak", "declining", "dormant", "expired", "recurring"] }, "level": { "enum": ["high", "medium", "low"] }, "desk": { "type": "array", "description": "One desk section. All five share this row shape.", "items": { "type": "object", "additionalProperties": false, "required": ["title", "note", "source", "why", "url"], "properties": { "title": { "type": "string" }, "note": { "type": "string", "description": "What to look at, or what the thing does. Empty when the rendered row carried no such line." }, "source": { "type": "string", "description": "Where it came from — the publisher, not the aggregator or search engine that surfaced it. Empty when the rendered row carried no source line." }, "why": { "type": "string", "description": "Why it was picked. Only the tools desk writes this; empty elsewhere." }, "url": { "type": "string", "description": "Source link, or the empty string. NOT `format: uri` for that reason." } } } }, "trend": { "type": "object", "additionalProperties": false, "required": [ "canonical_id", "name", "category", "lifecycle_stage", "momentum", "impact_level", "confidence_level", "summary", "attention_driver", "evidence", "first_detected", "last_signal_update" ], "properties": { "canonical_id": { "type": "string", "description": "Identity across runs. The SAME trend keeps this id as it moves through lifecycle stages, so a reader can reconstruct its history from the archive. Prefer this over `run_trend_id` for any cross-day join." }, "run_trend_id": { "type": "string", "description": "Human-readable id minted by this run. Not stable across runs." }, "name": { "type": "string" }, "category": { "type": "string", "description": "Taxonomy bucket, e.g. safety_and_governance. Vocabulary is in taxonomy.json." }, "primary_region": { "type": "string" }, "secondary_regions": { "type": "array", "items": { "type": "string" } }, "lifecycle_stage": { "$ref": "#/$defs/lifecycle" }, "momentum": { "enum": ["accelerating", "stable", "decelerating"] }, "impact_level": { "$ref": "#/$defs/level" }, "opportunity_level": { "$ref": "#/$defs/level" }, "confidence_level": { "$ref": "#/$defs/level", "description": "Model-assigned, driven mainly by the tier mix of the evidence behind the trend. Not a calibrated probability." }, "summary": { "type": "string", "description": "What happened." }, "attention_driver": { "type": "string", "description": "Why it is being noticed now." }, "so_what": { "type": ["string", "null"], "description": "Reader-facing implication. OPTIONAL and audience-dependent — omitted unless the run was rendered under a published, neutral lens. Never export a so_what written for a private audience." }, "evidence": { "type": "array", "minItems": 1, "items": { "$ref": "#/$defs/evidence" }, "description": "Every trend carries at least one citation. A trend with no evidence is not published." }, "first_detected": { "type": "string", "format": "date-time" }, "last_signal_update": { "type": "string", "format": "date-time" } } }, "evidence": { "type": "object", "additionalProperties": false, "required": ["url", "quote", "claim", "source_tier", "used_for"], "properties": { "content_hash": { "type": "string", "description": "Stable identity for the underlying item. This is the ONLY key that survives across the pipeline's internal artifacts — item ids and canonical URLs are both rewritten during normalization. Use `content_hash`, or `url`, to join evidence across days; never assume an id is stable." }, "url": { "type": "string", "format": "uri" }, "quote": { "type": "string", "description": "VERBATIM excerpt from the cited page. Not a paraphrase. Truncated, never edited." }, "claim": { "type": "string", "description": "The specific assertion this quote is being used to support." }, "source_tier": { "enum": ["primary_release", "independent_validation", "attention", "companion"] }, "used_for": { "enum": ["existence", "momentum", "impact", "sentiment", "creator_relevance"], "description": "Which part of the trend claim this citation supports. `existence` = the thing happened; `momentum` = it is accelerating; `impact` = it matters; `sentiment` = how it is being received. `creator_relevance` is rare on this feed — the synthesis prompt is shared across domains and offers it everywhere, so it can appear on an AI record even though it reads as a creator-economy category." } } } } }