What is a Guidance Credibility Index (GCI)?

A Guidance Credibility Index measures whether listed-company management delivered on quantified guidance — such as revenue bands, margins, volumes, or capex — against later reported actuals, with primary sources attached. It is a delivery score, not a sentiment score and not investment advice.

Is CiteAlpha GCI investment advice?

No. CiteAlpha, a product of Ocotillo Innovation Private Limited, ships GCI as factual research infrastructure for Indian equity desks. It does not provide Buy, Hold, or Sell recommendations or personalised investment advice.

What evidence does a citeable GCI point need?

A citeable point needs a dated guidance statement, a matched actual for the same metric and period, an outcome label (met, exceeded, missed, dropped, or pending), and a path back to NSE, BSE, or IR documents.

How is CiteAlpha GCI different from sentiment analysis?

Sentiment analysis tracks tone and narrative. GCI tracks whether quantified guidance was delivered. Desks should keep them in separate lines of a memo so tone never overrides the guidance-versus-actuals table.

Who is CiteAlpha built for?

Indian equity desks — buy-side and sell-side research teams that need evidence-linked management delivery history beside their market terminal. Pilots start with Sensex-depth hand-labeled evidence.

Help & Glossary for GCI Workflows — CiteAlpha

CiteAlpha Help explains Guidance Credibility Index vocabulary, outcome labels, and desk workflows for Indian equity research. Factual research product — not investment advice.

Start with the GCI Screener, About, or the canonical GCI definition. Contact sales@citealpha.com for pilots.

Where to work

GCI Screener: Universe screen for Sensex names: GCI level, Δ GCI (YoY/QoQ/PoP), quality badge, peers, and alerts. Open a name for the evidence dossier and charts.
Analyst Workbench (review & export workspace): Review workspace for pilot and paid desks: approve extracted guidance, check evidence, export point-in-time data, and build reports. Requires a pilot or paid seat.
Filing Search: One company at a time: search its filings and transcripts, ask cite-only questions, and see a snapshot of guidance vs reported numbers. Complements GCI. Not a price feed.
Disclosure Explorer: Compare disclosures across Indian companies: search, cite-only Ask, boards, compare grid, and desk agents over public IR and labeled evidence. Complements GCI — not a price terminal.
CiteAlpha Score: GCI 0–100 plus peer delivery benchmarks. The brand wedge — management promises vs delivery.
CiteAlpha Radar: Feed of guidance changes, misses, drops, and withdrawals — for PMs, risk, and IR-watch. Not a news ticker.
CiteAlpha Ledger: Promise ledger: what was committed, by whom, when, and status. For compliance, credit, IR, and board packs.
CiteAlpha Data: Point-in-time guidance-outcome dataset and factor export. Backtestable series — no look-ahead.
Public Snapshot: Public, no-login snapshot of citeable GCI names. Evidence-linked scores, not recommendations.
Enterprise platform: Single contract for GCI Screener + evidence + API + facts import + Wordmap + vernacular + customer success. Guidance accountability only — keep your market terminal for prices.
Trust Center: Public security, legal, and citation posture for procurement. Honest runtime flags — not marketing claims.
Tier 1 foundation: Ingest → structure → search → cite. Auto IR corpus, period doc types (transcript/results/IR), and citeable citation ids. Without this, analytics stay experimental.

Language, identity & markets

Language selector: Header control for UI chrome (Indian + global languages). Persists to localStorage and account preferences when logged in. Does not change GCI math. Glossary remains EN for now.
Guest session: One-click ephemeral identity with preferences. Register to sync language, market, index, and watchlist across devices. Auth does not imply investment-advice entitlement.
Market: Exchange region (IN, US, JP, …). India remains the deep GCI path; other markets ship index constituent lists with honest quality badges.
Index: Flagship benchmark within a market (e.g. SENSEX, Nifty 50, S&P 500). Filters the universe table. Full GCI labeling is India-first.

Core ideas

GCI (Guidance Credibility Index): 0–100 score of how reliably management delivers quantified guidance. Built from evidence rows (band → actual → label). Not a buy/sell recommendation.
Guidance: A forward-looking quantified promise from management for a metric and period (e.g. FY25 revenue growth 8–10%).
Actual: The reported result for that same metric and period — used to judge whether guidance was delivered.
Band (guided low–high): The range management guided to. Landing inside the band (small tolerance) counts as met.
Evidence trail: Auditable table: period, metric, band, actual, Δ Actual, label, vs-guide delta, and source. Prefer Hand-labeled names when citing.
Citeable: Hand-labeled outcome with source URL + quote (+ bound doc). Provisional/demo scores are never citeable — reports exclude them.
Corpus status: gci_citeable = hand-labeled with sources; gci_available = scored/docs present; listed_not_in_corpus = exchange listing only.
Promise thread: Restatement history of one guidance commitment: first stated, then raised / lowered / reiterated across calls, until it resolves (met / missed / exceeded / dropped). Quiet walk-backs show up here.
GCI parameter: A catalog metric management quantified for a period (~16 core + sector). Only these enter GCI — see Desk → Parameters charts.
Source policy: GCI uses text: transcripts, filings, IR, PPT text, ASR→transcript, reported actuals. Raw audio/video, technicals, and shenanigans do not score GCI.

Outcome labels

Met: Actual landed in (or near) the guided band. Full contribution (~100 before confidence).
Exceeded: Actual beat above the band. Default scorer v4: a small beat scores close to 100; a beat far above the band still means the guidance was off, so points decay with distance but never below 60. A miss the same distance below always scores lower (γ=1.4). v3 let beats decay toward 0; legacy v2 floors beats ≥85.
Missed: Actual below the band. Score decays with shortfall; large misses can go to 0.
Dropped: Management stopped reiterating guidance. Default scorer v4 (as in v3) excludes the row from the average and applies a company −15 withdrawal deduction. Legacy v2 scored dropped ≈35. Distinct from a miss.
Pending: Period still open or no actual yet. Excluded from company GCI until closed.
Pending guidance cite: The reported actual is cited, but the original guidance filing (URL, quote, date) is not yet on the row. Excluded from the company GCI until both sides are cited.

How the score is built

Confidence: Extraction/match certainty (0.5–1.0). Lower confidence pulls the outcome score toward 50.
Delta % (vs guide): Relative gap between actual and guided midpoint (or band edge) for that outcome.
Contribution score: That outcome’s 0–100 score after label rules and confidence. Company GCI averages these.
Metric: Catalog name for the guided measure (for example revenue growth). Unknown metrics are rejected on import.
Period: Fiscal window the guidance applies to (e.g. FY24, Q2FY25).

Change trends & history

MoM / QoQ / YoY change: Incremental % vs the prior comparable period. On desks, change trends usually matter more than absolute levels alone.
GCI trend: Period-by-period company GCI with a line chart — shows credibility improving or eroding. Read level and Δ together.
PIT history: Point-in-time GCI as of past dates for backtests — no look-ahead. Shown as a table and path chart with Δ vs prior.
Charts: Line, bar, and donut visuals for GCI path, metric Δ, parameter usage, and Wordmap. Levels stay in tables; charts highlight change.
Granger precedence: LASSO variable selection then Granger F-test on PIT series (≥12 quarters). Predictive precedence ≠ causation; not a forecast or investment advice.

Workbench fields

Peer rank: Rank of this company’s GCI among Sensex peers in the same sector (1 = best).
Sector avg: Average of company GCI scores in the sector. Each company is scored on its own guidance. The average is not a sector adjustment.
Alerts: Flagged events such as repeated misses or dropped guidance that need analyst attention.
Data quality: Hand-labeled = guidance whose quotes were found on the cited filings; the only rows that carry a GCI. Demo = sample data for walkthroughs, not scored. Listing = NSE/BSE master, not yet scored.
NSE / BSE listing: Real exchange equity master row. GCI stays blank until guidance vs actuals are labeled — we never invent actuals.
NSE_ALL / BSE_ALL: Full India equity masters for navigation. Only hand-labeled companies carry a GCI; every other listing shows Not yet scored until both quotes are found on the cited filings.
Source: Where the guidance came from (transcript, filing, IR) so every outcome stays auditable.
Review: Analyst Accept keeps the outcome; Reject marks it for exclusion/follow-up.
Extract: Pulls candidate guidance statements from transcript text into a needs-review queue before they enter GCI.
Wordmap (context): Entity vs industry tone themes for context only. Not part of GCI math.
Facts import: Paste catalog-aligned facts JSON and merge into a company trail. Not a live vendor connector.
Labeling priority queue: Enterprise backlog for hand-label priority. Queues a company for analyst labeling — does not invent actuals.
Customer success: Named customer-success contact and quarterly reviews — included on Enterprise. The Workbench shows your pilot organisation.
Label: Outcome class: met, exceeded, missed, dropped, or pending.